Cloud computing

Week-1
1. Introduction to Computing and Cloud
 Definition (ACM, 2005): Computing involves goal-oriented activities like hardware/software
design, information management, intelligent systems, entertainment, and communication.
 Evolution:
o Centralized Computing → Distributed Computing → Grid Computing → Cluster
Computing → Utility Computing → Cloud Computing.
 Distributed Systems: Multiple autonomous entities with local memory, connected via
message passing. Benefits: fault tolerance, scalability, load sharing, robustness, resource
sharing. Examples: Internet, ATMs, intranets.
2. Grid and Cluster Computing
 Grid Computing: Virtualizes distributed resources (processing, storage, bandwidth) to appear
as a single system.
o Types: Computational, Data, Collaboration, Network, Utility Grids.
o Key use: Large-scale scientific/engineering problems, multi-institution collaborations.
 Cluster Computing: Group of interconnected stand-alone computers working as one
integrated resource.
o Types: High availability, load balancing, parallel clusters.
o Benefits: High performance, scalability, fault tolerance, redundancy.
3. Utility Computing
 Concept of pay-per-use model (like electricity/water).
 Characteristics: On-demand, outsourcing, virtualization, automated provisioning, web service
delivery.
 Payment Models: Flat rate, subscription, metered, tiered, pay-as-you-go.
 Risks: Data security, SLA issues, dependency on providers.
4. Cloud Computing Basics (NIST Definition)
“Cloud computing is a model for enabling convenient, on-demand network access to a shared pool of
configurable computing resources… that can be rapidly provisioned and released with minimal
effort.”
 Essential Characteristics:
o On-demand self-service
o Broad network access
o Resource pooling
o Rapid elasticity
o Measured service
 Common Traits: Massive scale, virtualization, low-cost software, geographic distribution,
service orientation, resilience.
5. Cloud Service Models (XaaS)
 SaaS: Software delivered over internet (Google Docs, Salesforce CRM).
 PaaS: Platform for app development & deployment (Google App Engine, Azure).
 IaaS: Infrastructure resources on demand (AWS EC2, S3).
 Other “as a Service”: BPaaS, DaaS, BaaS, IDaaS, SECaaS, etc.
6. Cloud Deployment Models
 Private Cloud: Exclusive to one organization.
 Public Cloud: Open to public (Google Docs).
 Community Cloud: Shared by several orgs with common goals.
 Hybrid Cloud: Mix of two/more clouds (cloud bursting).
7. Virtualization in Cloud
 Virtual Machines (VMs) enable abstraction of physical machines.
 Hypervisors (VMware, Xen, KVM) manage VMs.
  Benefits: Hardware utilization, migration, testing, backup, cost efficiency, legacy system
support.
8. Advantages of Cloud Computing
 Cost reduction (hardware, software, power).
 High scalability & performance.
 Instant software updates.
 Universal access, collaboration, device independence.
 Reliability & unlimited storage.
9. Disadvantages of Cloud Computing
 Dependency on internet speed & connectivity.
 Limited features vs. desktop apps.
 Security & privacy risks.
 Vendor lock-in & interoperability issues.
 Performance issues for HPC (MPI/OpenMP).
10. Economics of Cloud
 Public clouds reduce IT costs vs. private data centers.
 Cost ratio shows cloud more efficient than in-house servers.
 PaaS vs. IaaS: PaaS auto-scales with demand.
 SaaS reduces software licensing costs.

✅ 50 Most Important MCQs with Answers

Basics of Cloud
1. Cloud computing is primarily based on which concept?
a) Centralized computing
b) Utility computing
c) Real-time computing
d) Parallel computing
Ans: b
2. Which organization defined the most accepted model of Cloud Computing?
a) ACM
b) NIST
c) IEEE
d) ISO
Ans: b
3. Which of the following is NOT an essential characteristic of cloud computing?
a) On-demand self-service
b) Broad network access
c) Resource pooling
d) Manual resource provisioning
Ans: d
4. Cloud systems measure usage via:
a) Rapid elasticity
b) Metering capability
c) Internet speed
d) Client OS
Ans: b
5. Which of these is a deployment model?
a) IaaS
b) PaaS
c) SaaS
d) Private Cloud
Ans: d

Distributed, Grid & Cluster

 6. Which is an example of distributed computing?
a) ATM Networks
b) Power Grids
c) LAN Switches
d) Single CPU Systems
Ans: a
7. Main property of distributed systems:
a) Centralized memory
b) Fault tolerance
c) High dependency on single node
d) Expensive to expand
Ans: b
8. Grid computing virtualizes which resources?
a) Storage, Processing, Bandwidth
b) Only Storage
c) Only CPU cycles
d) Only Network
Ans: a
9. Cluster computing improves:
a) Cost & Reliability
b) Only Cost
c) Only Reliability
d) None
Ans: a
10. Which cluster type balances workload?
a) Failover Cluster
b) Load Balancing Cluster
c) Parallel Cluster
d) Backup Cluster
Ans: b

Cloud Models
11. In SaaS, user controls:
a) Application code
b) Middleware
c) Infrastructure
d) None
Ans: d
12. Example of PaaS:
a) Gmail
b) Google App Engine
c) AWS EC2
d) Dropbox
Ans: b
13. In IaaS, user can:
a) Control OS & apps
b) Control hardware design
c) Control physical servers
d) Modify provider APIs
Ans: a
14. Community Cloud is shared by:
a) General public
b) One org
c) Multiple orgs with common goal
 d) None
Ans: c
15. Hybrid Cloud combines:
a) SaaS + IaaS
b) Public + Private/Community
c) Grid + Cluster
d) VPN + LAN
Ans: b

Virtualization & Networking

16. Software that manages VMs:
a) Hypervisor
b) Compiler
c) Emulator
d) Cache manager
Ans: a
17. VMware, Xen, KVM are examples of:
a) Cloud APIs
b) Hypervisors
c) SaaS Platforms
d) Middleware
Ans: b
18. Virtualization advantage:
a) Hardware dependency
b) Legacy support
c) Manual provisioning
d) Higher physical servers needed
Ans: b
19. Which network tools help in cloud?
a) OpenSSH, OpenVPN
b) MS Word, Excel
c) Hadoop, Spark
d) Oracle DB
Ans: a
20. NFV stands for:
a) Network Function Virtualization
b) New File Version
c) Network Fast Verification
d) None
Ans: a

Advantages & Disadvantages

21. Key advantage of cloud:
a) Higher local storage
b) Lower cost, scalability
c) Offline operation
d) Complete privacy
Ans: b
22. Main disadvantage of cloud:
a) Needs Internet connectivity
b) No scalability
c) Requires big servers locally
d) Cannot run on mobiles
Ans: a
 23. Cloud ensures unlimited storage via:
a) Hard disk expansion
b) Remote pooled resources
c) USB drives
d) RAID disks only
Ans: b
24. Data in cloud is prone to:
a) Redundancy only
b) Loss & security risks
c) Always 100% safe
d) Local OS issues only
Ans: b
25. Cloud is not ideal for:
a) Tax/Billing apps
b) Real-time HPC with MPI/OpenMP
c) CRM
d) Storage
Ans: b

Economics & Business

26. Cloud reduces IT cost mainly via:
a) Buying more hardware
b) Pay-per-use
c) No SLAs
d) Local software licenses
Ans: b
27. Public clouds provide:
a) High scalability & SLA (~99%)
b) Zero security
c) Unlimited free hardware
d) On-premise only
Ans: a
28. SMBs benefit from cloud because:
a) High upfront cost
b) Reduced need for infra
c) Local-only access
d) Only large enterprises use it
Ans: b
29. Which service is best for startups with limited budget?
a) On-premise servers
b) Renting cloud infra (IaaS/PaaS)
c) Mainframe systems
d) Clusters only
Ans: b
30. PaaS scales automatically depending on:
a) Pre-purchased servers
b) Usage load
c) Local IT staff
d) Vendor lock-in
Ans: b

More Applied MCQs (31–50, condensed)

31. Cloud storage example → Amazon S3
32. SaaS CRM example → Salesforce
 33. Example of Cloud Bursting → Hybrid Cloud
34. Measured service ensures → Usage transparency
35. Virtual workspaces created via → VMs
36. Key risk in utility computing → Defining SLA
37. Google Docs is → SaaS
38. Azure is → PaaS
39. AWS EC2 is → IaaS
40. Dropbox is → Cloud Storage SaaS
41. Cluster networks usually → Low latency
42. Grid computing analogy → Electrical Power Grid
43. Cloud feature that saves data from PC crash → Reliability
44. Example of Community Cloud → Govt sector shared cloud
45. Advantage of virtualization → Migration of VMs
46. Vendor lock-in problem in → Cloud APIs differences
47. Main driver for adopting cloud → Reduce IT cost
48. Device independence means → Access from any device
49. SaaS avoids → Manual upgrades
50. Pay-as-you-go is → Utility pricing model
1. Introduction to Computing & Cloud
 Computing (ACM, 2005): Any activity that needs computers – like designing
hardware/software, managing data, doing research, AI, entertainment, or communication.
 Evolution Path:
Centralized → Distributed → Grid → Cluster → Utility → Cloud Computing.
 Distributed Systems: Many independent computers connected together, sharing work.
o Benefits: Fault tolerance, scalability, load sharing, robustness, resource sharing.
o Examples: Internet, ATMs, intranet systems.

2. Grid & Cluster Computing

 Grid Computing: Combines unused resources (CPU, storage, bandwidth) to look like one
supercomputer.
o Types: Computational, Data, Collaboration, Network, Utility.
o Use: Solving big science/engineering problems (e.g., CERN data analysis).
 Cluster Computing: A set of computers connected to act as one.
o Types: High availability (failover), load balancing, parallel clusters.
o Benefits: High performance, scalability, reliability, and fault tolerance.

3. Utility Computing
 Works like electricity/water billing → pay only for what you use.
 Features: On-demand, outsourcing, virtualization, automated setup.
 Payment Models: Flat rate, subscription, metered, tiered, pay-as-you-go.
 Risks: Data security, dependency on providers, SLA (service level agreement) issues.

4. Cloud Computing Basics (NIST Definition)

"On-demand, convenient access to shared computing resources (servers, storage, apps, services) that
can be quickly scaled up or down."
 Key Features:
o Self-service on demand
o Accessible anywhere via network
o Shared pooled resources
o Elastic scaling (up/down)
o Pay-per-use (measured service)
 Other Traits: Low cost, virtualized, distributed, secure, large-scale.

5. Cloud Service Models (XaaS)

  SaaS (Software as a Service): Ready-to-use software (e.g., Google Docs, Salesforce).
 PaaS (Platform as a Service): Platform to build & deploy apps (e.g., Azure, Google App
Engine).
 IaaS (Infrastructure as a Service): Virtual servers, storage, networks (e.g., AWS EC2, S3).
 Others: BPaaS, DaaS, BaaS, IDaaS, SECaaS etc.

6. Cloud Deployment Models

 Private Cloud: For one organization only.
 Public Cloud: Open for everyone (e.g., Google Docs).
 Community Cloud: Shared by a group with common needs.
 Hybrid Cloud: Mix of 2+ types (e.g., cloud bursting for load balancing).

7. Virtualization in Cloud
 Virtual Machines (VMs): Simulate computers on one physical machine.
 Hypervisors: Software to manage VMs (e.g., VMware, Xen, KVM).
 Benefits: Better hardware use, backup & migration, cost savings, legacy system support.

8. Advantages of Cloud Computing

 Lower cost (no need for powerful personal computers).
 High scalability and performance.
 Automatic software updates.
 Access from anywhere (device independence).
 Collaboration is easier.
 Reliable data storage (virtually unlimited).

9. Disadvantages of Cloud Computing

 Needs constant internet.
 Doesn’t work well on slow connections.
 Fewer features than some desktop apps.
 Data security and privacy issues.
 Vendor lock-in (hard to switch providers).
 Not ideal for high-performance scientific computing (HPC).

10. Economics of Cloud

 Public clouds: Cheaper than private data centers.
 Cost efficiency: Cloud servers more effective than in-house ones.
 PaaS vs IaaS: PaaS automatically scales apps with demand.
 SaaS: Saves on costly software licenses.
 Week-2
Cloud Computing Architecture & Deployment Models
Cloud computing provides shared, scalable computing resources as services. The deployment
models define how these resources are managed:
 Public Cloud: Infrastructure open to general public use. Examples: AWS EC2, Google App
Engine, Microsoft Azure. Offers elasticity and low cost but has security risks due to multi-
tenancy and limited control.
 Private Cloud: Used exclusively by one organization. Offers more control and security but
higher setup costs.
 Community Cloud: Shared by several organizations with similar goals (e.g., universities,
banks). Offers shared costs and policies.
 Hybrid Cloud: Combination of two or more clouds (public, private, community) linked for
portability and flexibility.
Each model varies in security, cost, scalability, and data control. Hybrid models offer flexibility but
add complexity.

2️⃣ Virtualization in Cloud Computing

Virtualization enables multiple OS instances on a single physical machine using a hypervisor
(Virtual Machine Monitor).
 IaaS (Infrastructure as a Service): Provides VMs, storage, and networking (e.g., AWS EC2).
 Cloud Manager: Handles user authentication, VM creation, and resource allocation.
 Cluster Manager: Manages groups of computers and allocates workloads.
 Computer Manager: Operates on each physical node to control VMs.
Types of Virtualization:
 Full Virtualization: Hardware fully emulated (QEMU, Bochs). OS unaware of virtualization.
 Para-Virtualization: Guest OS modified to work with hypervisor. Faster than full.
 Hardware-assisted Virtualization: Uses CPU support (Intel VT-x, AMD-V).
Advantages: Resource efficiency, isolation, scalability.
Challenges: Performance overhead, security management.

3️⃣ XML (Extensible Markup Language)

XML is a text-based, platform-independent format for data representation and transfer. It defines
custom tags to describe data.
Key Concepts:
 DTD & Schema: Define structure and data types of XML.
 Namespaces: Prevent naming conflicts in combined documents.
 XML Parsers:
o SAX (Simple API for XML): Event-based, fast, low memory.
o DOM (Document Object Model): Tree-based, editable in-memory.
o JDOM: Java-specific, simplifies DOM usage.
 XSLT (Extensible Stylesheet Language Transformation): Transforms XML into other formats
(HTML, XML).
XML supports data exchange across heterogeneous systems and forms the backbone of Web
Services (SOAP, WSDL, UDDI).

4️⃣ Web Services & Service-Oriented Architecture (SOA)

Web Services are modular applications that can be described, published, located, and invoked over
a network using standard protocols.
They solve issues of interoperability, firewall traversal, and system complexity.
Core Technologies:
 SOAP (Simple Object Access Protocol): XML-based communication between applications
 over HTTP/SMTP.
 WSDL (Web Service Description Language): Defines what a service does, message formats,
and access protocols.
 UDDI (Universal Description, Discovery, Integration): Registry for publishing and
discovering web services.
SOA (Service-Oriented Architecture): Framework with three roles —
 Service Provider: Creates and hosts services.
 Service Broker: Maintains service directory (UDDI).
 Service Requestor: Finds and invokes services.
Advantages: Platform-independent, reusable, interoperable, and scalable systems for distributed
applications.

NPTEL Cloud Computing – Week 2 Summary (Easy Notes)

1️⃣ Cloud Computing Deployment Models
Cloud deployment models define how and where the cloud infrastructure is managed and who can
access it.
Model Description Example Key Points
Services available to everyone AWS EC2, Azure, Low cost, high scalability, less
Public Cloud
over the Internet Google Cloud control, security risk
Used by a single organization VMware, Amazon High control & security, high
Private Cloud
(on-site or outsourced) VPC cost
Community Shared by several organizations Govt. or university
Shared cost & policies
Cloud with common goals clouds
Combines 2 or more models VMware vCloud, Flexibility, complexity in
Hybrid Cloud
(public, private, community) Azure Hybrid management
Key Idea:
Public = Open use, Private = Secure use, Community = Shared use, Hybrid = Mixed use.

2️⃣ Virtualization – the Heart of Cloud

Virtualization allows multiple operating systems to run on one physical machine using a hypervisor
(VMM).
It separates software from hardware → better resource use, isolation, and scalability.
Types of Virtualization
 Full Virtualization: Hardware fully emulated (e.g., QEMU). Slower but unmodified OS
works.
 Para-Virtualization: OS modified to talk directly with hypervisor → faster.
 Hardware-Assisted: CPU provides virtualization support (Intel VT-x, AMD-V).
Infrastructure Layers (IaaS)
 Cloud Manager – user authentication & resource control
 Cluster Manager – controls groups of computers
 Computer Manager – runs on each node, starts/stops VMs

3️⃣ XML – Data Language of the Cloud

XML (eXtensible Markup Language) is a universal text format to store and share data.
You can create your own tags like <student><name>Prince</name></student>.
Why XML?
 Platform-independent
 Self-describing (data + structure)
 Easy to exchange between systems
Validation Tools
 DTD – Defines structure & allowed elements.
 Schema (XSD) – Stronger version of DTD; defines data types.
  Namespaces – Avoid name clashes when mixing vocabularies.
Parsers
 SAX – Event-based, very fast, low memory.
 DOM – Tree-based, editable, slower.
 JDOM – Java-friendly version of DOM.
 XSLT – Transforms XML into another format (like HTML).

4️⃣ Web Services & Service-Oriented Architecture (SOA)

Web Services = Software components that communicate via the Internet using XML messages.
They solve interoperability issues between applications on different platforms.
Core Technologies
 SOAP: XML-based messaging protocol over HTTP/SMTP.
 WSDL: Describes what the service does and how to access it.
 UDDI: Directory for publishing and discovering web services.
SOA Roles
 Service Provider – creates and hosts service
 Service Broker/Registry – stores service details (UDDI)
 Service Requestor – finds and uses services
Benefits: Platform independence, reusability, loose coupling, scalability.

🧩 50 Most Important MCQs

1. Cloud infrastructure open to the public is called —
a) Private Cloud b) Public Cloud c) Hybrid Cloud d) Community Cloud
✅ Ans: b) Public Cloud
2. Which model offers maximum control?
a) Hybrid b) Public c) Private d) Community
✅ c) Private
3. Community cloud is for —
a) General public b) Specific organization c) Shared interests group d) Cloud provider
✅ c) Shared interests group
4. Hybrid cloud connects —
a) Two or more clouds b) Two servers c) Only private clouds d) Databases
✅ a) Two or more clouds
5. Example of public cloud —
a) Amazon EC2 b) VMware c) Eucalyptus d) OpenNebula
✅ a) Amazon EC2
6. Which cloud has highest upfront cost?
a) Public b) Private (on-site) c) Community d) Hybrid
✅ b) Private (on-site)
7. Main risk in public cloud —
a) High cost b) Multi-tenancy c) Less elasticity d) Low performance
✅ b) Multi-tenancy
8. “Elasticity” in cloud means —
a) Scalable resources b) Low security c) Fixed resources d) Free storage
✅ a) Scalable resources
9. In IaaS, subscriber rents —
a) Applications b) Operating System c) Virtual Machines and storage d) Network
protocols
✅ c) Virtual Machines and storage
10. Hypervisor is also called —
a) VMM b) API c) Container d) Driver
✅ a) VMM
11. IaaS provider manages —
a) Hardware & Hypervisor b) Apps c) Guest OS d) Middleware
 ✅ a) Hardware & Hypervisor
12. Guest OS runs inside —
a) Virtual Machine b) Hypervisor c) Kernel d) Cloud Manager
✅ a) Virtual Machine
13. Full virtualization emulates —
a) Software b) Hardware c) Network d) User interface
✅ b) Hardware
14. Para-virtualization requires —
a) Modified OS b) Extra hardware c) Binary translation d) Dual boot
✅ a) Modified OS
15. Hardware-assisted virtualization uses —
a) Intel VT/AMD-V b) BIOS c) VMware Tools d) Containers
✅ a) Intel VT/AMD-V
16. XML is —
a) Programming language b) Markup language c) Database d) Protocol
✅ b) Markup language
17. XML helps in —
a) Web page design b) Data representation and exchange c) Animation d) Encryption
✅ b) Data representation and exchange
18. A well-formed XML means —
a) Has DTD b) Has root element and correct syntax c) Has schema d) Has namespace
✅ b) Has root element and correct syntax
19. Valid XML document —
a) Only well-formed b) Matches DTD/Schema c) Compressed d) Encrypted
✅ b) Matches DTD/Schema
20. Schema is written in —
a) HTML b) XML c) Java d) C++
✅ b) XML
21. XML Namespace avoids —
a) Duplicate names b) White spaces c) Data loss d) Encoding errors
✅ a) Duplicate names
22. SAX parser is —
a) Tree-based b) Event-based c) Hybrid d) DOM-based
✅ b) Event-based
23. DOM parser stores data as —
a) Binary b) Text stream c) Tree structure d) JSON
✅ c) Tree structure
24. JDOM works in —
a) Java b) Python c) C# d) C++
✅ a) Java
25. XSLT is used for —
a) Encryption b) Transformation c) Parsing d) Validation
✅ b) Transformation
26. Web Service communicates via —
a) Binary files b) XML messages c) PDFs d) HTML pages
✅ b) XML messages
27. SOAP stands for —
a) Simple Object Access Protocol b) Secure Object Application Protocol c) System
Online Access Platform d) None
✅ a) Simple Object Access Protocol
28. SOAP is based on —
a) HTTP/SMTP b) FTP c) SSH d) SNMP
✅ a) HTTP/SMTP
29. WSDL defines —
 a) Service interface and operations b) Hardware config c) Security policy d) Database
schema
✅ a) Service interface and operations
30. UDDI is used for —
a) Service Discovery b) Data Storage c) Encryption d) Testing
✅ a) Service Discovery
31. Web Service architecture roles are —
a) Provider, Broker, Requestor b) Server, Client, Admin c) User, Vendor, Host d) OS,
App, DB
✅ a) Provider, Broker, Requestor
32. SOA stands for —
a) Service-Oriented Architecture b) System Open Application c) Software Operating
Architecture d) None
✅ a) Service-Oriented Architecture
33. Main goal of SOA —
a) Loose coupling of services b) Tight integration c) Single platform d) Hardware
virtualization
✅ a) Loose coupling of services
34. SOAP messages are written in —
a) Binary b) XML c) HTML d) JSON
✅ b) XML
35. WSDL document uses —
a) XML b) C++ c) Java d) YAML
✅ a) XML
36. IaaS cost measured by —
a) CPU & storage usage b) Fixed license c) User count d) Monthly fee only
✅ a) CPU & storage usage
37. Virtualization increases —
a) Resource utilization b) Hardware cost c) Downtime d) Complexity only
✅ a) Resource utilization
38. Cluster Manager is responsible for —
a) Managing VM clusters b) User login c) Data security d) Billing
✅ a) Managing VM clusters
39. Cloud Manager handles —
a) Authentication & resource allocation b) CPU scheduling c) Backup d) Networking
only
✅ a) Authentication & resource allocation
40. Hypervisor provides —
a) Virtual machines b) Databases c) Web pages d) Containers only
✅ a) Virtual machines
41. XML vs HTML – main difference —
a) XML describes data, HTML displays data ✅
42. SOAP is —
a) Stateless protocol ✅
43. UDDI registry stores —
a) WSDL service information ✅
44. Hybrid cloud advantage —
a) Flexibility ✅
45. Private cloud disadvantage —
a) High initial cost ✅
46. Public cloud depends on —
a) Internet infrastructure (DNS, routers) ✅
47. XML parsers check —
a) Syntax & validity ✅
 48. XML messaging uses —
a) HTTP, SMTP ✅
49. Service-oriented architecture improves —
a) Reusability & interoperability ✅
50. Cloud computing enables —
a) On-demand self-service access ✅

Week-3
Service Level Agreements, Economics, and Data Management in Cloud Computing
I. Service Level Agreements (SLA) and Service Level Objectives (SLO)
A
Service Level Agreement (SLA) is a formal contract between a Service Provider (SP) and a Service
Consumer (SC) that serves as the foundation of the consumer's trust. Its core purpose is to define a
formal basis for the performance and availability that the SP guarantees to deliver. The SLA contains
Service Level Objectives (SLOs), which are objectively measurable conditions for the service. SLOs
encompass various Quality of Service (QoS) parameters, such as availability, throughput, response
time, and penalties.
Key contents of an SLA include a specific definition of services, the responsibilities of the provider
and consumer, a set of metrics to measure adherence, an auditing mechanism, and remedies for
non-satisfaction. SLAs can be
Off-the-shelf (non-negotiable/Direct SLA), which are common in present-day state-of-the-art clouds
and are less conducive for mission-critical data, or Negotiable, which may involve external agents.
Cloud SLAs differ from traditional Web Service SLAs, particularly in QoS parameters, where Cloud
SLAs emphasize security, privacy, and trust. Automation is crucial for Cloud SLA negotiation,
provisioning, and monitoring due to the highly dynamic and scalable nature of cloud service
consumption.
Monitoring and Auditability are vital.
Key Performance Indicators (KPIs) are low-level resource metrics (e.g., downtime, inbytes, packet
size) that are aggregated or converted to form high-level SLOs. To eliminate the conflict of interest, a
neutral third-party organization is often the best solution for monitoring performance. Essential
metrics include
Availability (uptime percentage), Durability (likelihood of data loss), Elasticity (ability to scale, with
limits clearly stated), and Resolution Time (time between detection and resolution of a problem).

II. Cloud Computing Economics: Value Propositions

Cloud properties offer significant economic advantages:
 Common Infrastructure: This property, driven by statistical multiplexing, enables Economies
of Scale (reduced overhead, volume purchasing) and Statistics of Scale.
Statistics of Scale allow for lower cost per delivered resource by multiplexing variable demands,
leading to higher resource utilization compared to unconsolidated workloads. The
Coefficient of Variation (CV), the ratio of standard deviation (σ) to the mean (μ), measures the
"smoothness" of demand. Aggregating

n independent demands reduces the CV by 1/n , which significantly minimizes the penalty of
insufficient or excess resources.
 Utility Pricing: This pay-per-use model is economical when demand is highly variable or spiky
(e.g., flash floods, holiday shopping). Owning resources is cheaper than renting only if the
Utility Premium (U) is less than the ratio of Peak Demand (P) to Average Demand (A) (U<P/A). A
hybrid model, combining owned baseline resources and rented cloud resources for peaks, is often
the best solution.
 On-Demand Resources: This elasticity ensures scalable resources are provisioned and de-
provisioned without delay. For an exponential demand
D(t)=et, any fixed provisioning interval will fall exponentially behind, leading to exponentially
increasing penalty costs.

III. Cloud Data Management and Distributed Systems

Relational databases have been largely supplanted in massive-scale web processing by distributed,
fault-tolerant file systems and NoSQL stores, such as
Google File System (GFS) and BigTable.
 Distributed File Systems (GFS/HDFS): GFS manages large files across a distributed cluster of
commodity servers, ensuring fault tolerance and supporting parallel operations. Its
architecture consists of a single
Master (NameNode in HDFS) that controls the file namespace and Chunk servers (DataNodes in
HDFS) that store large files broken into chunks, typically 64 MB in size, replicated three times on
different physical racks.
 NoSQL/Structured Storage:
o BigTable: A distributed structured storage system built on GFS. It is a
sparse, persistent, multi-dimensional sorted map (key-value pairs) accessed by Row Key, Column
Key, and Timestamp. Data is split into row ranges called
tablets, each managed by a tablet server.
o Dynamo: Developed by Amazon, it is a simple <key, value> pair data store supporting
a large volume of concurrent, small updates. It uses a
ring architecture and the Quorum Protocol (Read replicas R + Write replicas W> Total replicas N) to
maintain eventual consistency. Write operations use distributed object
versioning with timestamps to capture update history and resolve conflicts.
 MapReduce: This is a programming model for massively parallel computations. It involves
two phases coordinated by a single master process:
1. Map Phase: Mappers read input, transform one set of key-value pairs to another,
sort results by key, and write them to a local file per reducer.
2. Reduce Phase: Reducers fetch partial computations from all mappers, group results
by the same key, perform a function, and write the final results back to the global file
system (GFS).
 OpenStack Core Components: OpenStack is an open-source cloud operating system:
o Keystone: Provides identity, token, and service catalog management (Authorization).
o Nova: Manages compute resources (VM instantiation, provisioning).
o Neutron: Manages network connectivity (IP, gateway, L2 connectivity).
o Glance: Provides a catalog and repository for virtual machine disk images.
o Cinder: Manages persistent Block Storage for VMs.
o Swift: Manages scalable, accessible from anywhere Object Storage for files and VM
images.
o Ephemeral Storage is local to the VM and managed by Nova, persisting only until the
VM is terminated.

50 Most Important NPTEL Exam Questions (MCQ)

Section A: Service Level Agreements (SLA) and QoS
Q1. What is the primary function of a Service Level Agreement (SLA)? A. To define the pricing model
for cloud resources. B. To create a formal contract defining guaranteed performance and availability.
C. To manage the physical infrastructure of the cloud provider. D. To audit the consumer's
compliance with payment schedules. Correct Answer: B
Q2. Objectively measurable conditions for a service defined within an SLA are known as: A. Key
Performance Indicators (KPIs). B. Quality of Service (QoS) metrics. C. Service Level Objectives (SLOs).
D. Web Service Agreements (WS-Agreement). Correct Answer: C
Q3. Which of the following is typically a primary QoS parameter considered in Cloud Computing
SLAs, but less so in traditional Web Service SLAs? A. Response time. B. Availability. C. Security and
privacy.
D. Cost of service. Correct Answer: C
Q4. The XML-based language and protocol used for negotiating, establishing, and managing service
agreements at runtime is known as: A. WSLA. B. UDDI. C. WS-Agreement.
D. XML-SLA. Correct Answer: C
Q5. The type of SLA common in state-of-the-art clouds where the provider defines all criteria and it
is non-negotiable is called: A. Custom SLA. B. Direct SLA or Off-the-shelf SLA.
C. Third-party SLA. D. Dynamic SLA. Correct Answer: B
Q6. What is the suggested method to eliminate the conflict of interest in performance monitoring
between a cloud provider and consumer? A. The consumer should monitor performance only. B.
The provider should report outages at its sole discretion. C. A neutral third-party organization should
perform monitoring.
D. Monitoring should be done via a ticketing system. Correct Answer: C
Q7. Which metric represents "How quickly the service responds"? A. Reliability. B. Availability. C.
Throughput.
D. Durability. Correct Answer: C
Q8. The ability for a given cloud resource to grow infinitely, with limits clearly stated, is defined by
the metric: A. Agility. B. Load balancing. C. Linearity. D. Elasticity.
Correct Answer: D
Q9. The time period between the detection of a service problem and its resolution is defined as: A.
Transaction time. B. Resolution time.
C. Response time. D. Service-level violation rate. Correct Answer: B
Q10. Durability in an SLA context refers to: A. How quickly the provider responds to resource scale-
up/down. B. How likely the data is to be lost.
C. How often the service is available. D. How a system performs as the load increases. Correct
Answer: B
Section B: Cloud Computing Economics and Value
Q11. The term used when a service's performance remains constant as the load increases is: A.
Elasticity. B. Agility. C. Linearity.
D. Scalability. Correct Answer: C
Q12. The economic advantage of Common Infrastructure generated by pooled, standardized
resources with variable demand is primarily known as: A. Economies of Scale. B. Multi-tenant
advantage. C. Statistical Multiplexing or Statistics of Scale.
D. On-demand resources. Correct Answer: C
Q13. The ratio of the standard deviation (σ) to the absolute value of the mean (∣μ∣) of demand is
known as the: A. Variance. B. Correlation coefficient. C. Coefficient of Variation (CV).
D. Utility Premium. Correct Answer: C
Q14. Aggregating n independent demands reduces the Coefficient of Variation (CV) by a factor of:

A. 1/n. B. n . C.

1/n .
D. n2. Correct Answer: C
Q15. In Utility Pricing, cloud rental cost is cheaper than owning baseline resources when: A.
Average demand (A) is greater than Peak demand (P). B. Utility Premium (U) is greater than the ratio
of P to A. C. Utility Premium (
U) is less than the ratio of P to A.
D. Peak demand (P) is equal to Average demand (A). Correct Answer: C
Q16. The value of on-Demand Services is evident when demand is: A. Flat. B. Linear. C. Exponential
or spiky.
D. Highly correlated. Correct Answer: C
Q17. The key implication of Location-independence in cloud computing is: A. Reduction in cost of
computation. B. Ubiquitous availability meeting performance requirements.
C. Increased security through geographical separation. D. The necessity of a single central directory
(UDDI). Correct Answer: B
Q18. For perfectly correlated demands across n workloads, the Coefficient of Variation (CV): A.

Reduces by 1/n . B. Remains constant.

C. Increases linearly with n. D. Tends to zero. Correct Answer: B
Q19. The concept that cloud resources are provisioned and de-provisioned without delay or costs
associated with change is known as: A. Statistical multiplexing. B. Location-independence. C. Utility
pricing. D. On-Demand Resources.
Correct Answer: D
Q20. When is a hybrid model (owning some resources and renting cloud for others) most
beneficial? A. When the total demand is flat. B. When the ratio of peak to average demand is
extremely low. C. When demands are often highly spiky.
D. When network cost is zero. Correct Answer: C
Section C: Cloud Data Management, Storage, and MapReduce
Q21. Which type of database storage is optimal for write-oriented operations like transaction
processing applications, storing relational records on contiguous disk pages? A. Column-oriented
storage. B. Shared nothing storage. C. Row-oriented storage.
D. Shared disk storage. Correct Answer: C
Q22. Which parallel database architecture consists of a cluster of independent servers, each with
its own disk space, connected by a network? A. Shared memory. B. Shared disk. C. Shared nothing.
D. Symmetric Multi-Processing (SMP). Correct Answer: C
Q23. In Google File System (GFS) and Hadoop Distributed File System (HDFS), large files are broken
up into units called chunks or blocks, typically of what size? A. 4 KB. B. 1 MB. C. 64 MB.
D. 128 KB. Correct Answer: C
Q24. In the GFS architecture, which component controls the file namespace and maintains regular
communication with storage nodes? A. Chunk Server. B. Client Program. C. Master (NameNode in
HDFS).
D. Distributed File System. Correct Answer: C
Q25. GFS chunks are typically replicated how many times on different physical racks/network
segments for fault tolerance? A. One time. B. Two times. C. Three times.
D. Five times. Correct Answer: C
Q26. BigTable is a distributed structured storage system characterized as a: A. Relational database
with full SQL support. B. Hierarchical file system. C. Sparse, persistent, multi-dimensional sorted map
(key-value pairs).
D. Distributed block storage system. Correct Answer: C
Q27. Data in a BigTable cell is accessed by the triplet: A. Table ID, Partition Key, Version ID. B. Row
Key, Column Key, Timestamp.
C. Key, Value, TTL (Time to Live). D. File Path, Offset, Chunk Server ID. Correct Answer: B
Q28. The core data structure/model for Amazon's Dynamo is: A. A row-oriented table. B. A
relational schema. C. A simple <Key, Value> pair.
D. A column-oriented database. Correct Answer: C
Q29. In Dynamo, the condition (R+W)>N, where R is read replicas, W is write replicas, and N is total
replicas, ensures: A. Strong Consistency. B. Sequential Consistency. C. Quorum Consistency.
D. Read/Write Efficiency. Correct Answer: C
Q30. Google App Engine's Datastore stores all entities in one BigTable table and uses which
component to support efficient prefix and range queries on key values? A. Single-Property indexes
and 'Kind' indexes.
B. The Root Tablet. C. Column-oriented storage. D. Separate storage engines (e.g., Berkley DB).
Correct Answer: A
Q31. The first phase of the MapReduce programming model is the: A. Shuffle phase. B. Reduce
phase. C. Map phase.
D. Aggregate phase. Correct Answer: C
Q32. In the MapReduce model, if a mapper fails, the master reassigns the designated key-range for
re-execution because: A. The partial computation is written into GFS. B. The partial computation is
written to the mapper’s local file system.
C. The reduce phase is already completed. D. The master stores the partial computation. Correct
Answer: B
Q33. The type of parallel computing model best suited for compute-intensive scientific tasks is
traditionally the: A. Distributed-memory model. B. Shared-memory model.
C. Shared-nothing model. D. Shared-disk model. Correct Answer: B
Q34. The core idea behind MapReduce efficiency in large-scale data processing is: A. Increasing the
number of reads by each processor. B. Ensuring every read is useful and contributes to the final
answer.
C. Relying on shared-memory architecture. D. Avoiding all inter-process communication. Correct
Answer: B
Q35. In MapReduce, who informs the reducers where the partial computations have been stored
on the local files of the mappers? A. The Reducer itself. B. The Mapper. C. The Master.
D. The Global File System (GFS). Correct Answer: C
Section D: OpenStack Architecture and Cloud Storage
Q36. In OpenStack, the component responsible for virtual machine (VM) instantiation and
provisioning (compute resources) is: A. Keystone. B. Cinder. C. Nova.
D. Neutron. Correct Answer: C
Q37. Which OpenStack component is primarily responsible for managing network connectivity,
including configuring IP, gateway, and DNS names? A. Nova. B. Glance. C. Neutron.
D. Swift. Correct Answer: C
Q38. In OpenStack, Block Storage for VMs, which persists until specifically deleted by the user and
is accessible as a block device (e.g., /dev/vdc), is managed by: A. Swift. B. Glance. C. Keystone. D.
Cinder.
Correct Answer: D
Q39. Which OpenStack component provides the central service for identity, token, and service
catalog management (Authorization)? A. Nova. B. Keystone.
C. Horizon. D. Glance. Correct Answer: B
Q40. Which OpenStack storage type persists until VM termination, is accessible as a local file
system from within the VM, and is used to run the operating system or scratch space? A. Object
Storage. B. Ephemeral Storage.
C. Block Storage. D. Local Volume Storage. Correct Answer: B
Q41. Which OpenStack component provides a catalog and repository for virtual machine disk
images? A. Nova. B. Neutron. C. Glance.
D. Horizon. Correct Answer: C
Q42. Object Storage in OpenStack is managed by which component and is typically used for storing
files, including VM images, accessible from anywhere? A. Cinder. B. Swift.
C. Nova. D. Glance. Correct Answer: B
Q43. The OpenStack storage type that persists until specifically deleted by the user and is
accessible from anywhere is: A. Ephemeral Storage. B. Block Storage. C. Instance Storage. D. Object
Storage.
Correct Answer: D
Q44. In a typical OpenStack VM creation workflow, which component is consulted first after the
user logs into Horizon and initiates VM creation? A. Nova. B. Cinder. C. Keystone.
D. Neutron. Correct Answer: C
Q45. After Nova initiates provisioning and saves the state to the DB, the next component to
interact is typically the: A. Glance for image lookup. B. Nova Scheduler to find an appropriate host.
C. Cinder for block device. D. Neutron for networking. Correct Answer: B
Q46. What OpenStack component is responsible for retrieving the image URI looked up through
Glance? A. Nova Scheduler. B. Cinder. C. Swift.
D. Neutron. Correct Answer: C
Q47. If a cloud consumer powers off a VM, the SLA requirement for the provider to zero out the
memory and possibly disk platters is known as: A. Data Retention and Deletion. B. Privacy. C.
Hardware Erasure and Destruction.
D. Data Encryption. Correct Answer: C
Q48. Which common cloud provider service listed in the document explicitly defined no specific
SLA guarantees for availability? A. Amazon S3. B. Amazon EC2. C. Google App Engine. D. Amazon
SimpleDB.
Correct Answer: D
Q49. In the context of the economic value of utility pricing, what concept is often considered the
best solution when demands are highly spiky? A. Owning all peak resources. B. Using a hybrid model
(baseline ownership + cloud rental).
C. Avoiding network-attached storage. D. Using highly correlated demands. Correct Answer: B
Q50. The core principle of parallel efficiency states that if a task takes time T on a uniprocessor, it
should ideally take T/p if executed on p processors, due to which factor? A. Message
communication overhead. B. Need for synchronization. C. Imbalance in work distribution. D. Scalable
utilization of resources.
Correct Answer: D
summary make humanier free
You're looking for a more human-friendly, concise summary of the PDF content.
Here is a human-centric summary of the key concepts from the file, focusing on how Service Level
Agreements (SLA), Cloud Economics, and Data Management affect a user's experience with cloud
services.

Human-Friendly Cloud Computing Summary

1. The Cloud Contract: Service Level Agreements (SLA)
Think of an SLA as the "Cloud Promise." It's a legally binding contract between you (the consumer)
and the cloud provider (like Amazon or Google) that establishes trust.
 What it Guarantees (SLOs): Inside the SLA are Service Level Objectives (SLOs), which are the
specific, measurable guarantees. These aren't vague; they include objective metrics for:
o Availability: How much time the service is up (e.g., 99.999% uptime).
o Durability: The likelihood that your data will not be lost.
o Response Time: How fast the service delivers results.
o Security/Privacy: Crucial guarantees about protecting your data.
 The Bottom Line: If the cloud provider fails to meet an SLO (like a major outage), the SLA
outlines the penalties (remedies) they must provide, usually in the form of service credits. To
ensure fairness, the monitoring of these metrics is often best done by a neutral third party
rather than the provider themselves.

2. The Cloud's Value: Economics & Scale

The cloud is often cheaper than running your own servers due to two main reasons:
 1. Sharing is Saving (Statistics of Scale): Cloud providers pool the computing needs of
hundreds of thousands of different customers onto the same massive infrastructure. Because
not everyone needs peak resources at the exact same moment, the provider can efficiently
juggle the variable demands (like a spiky Black Friday rush versus a quiet Tuesday). By
aggregating these varying demands, they don't have to buy a massive amount of hardware
just to handle a few high peaks, which ultimately lowers the cost for everyone.
 2. Pay-Per-Use (Utility Pricing): The cloud works like your electricity bill—you only pay for
what you use (on-demand resources).
o This is incredibly valuable if your demand is "spiky" (sudden, unpredictable surges).
You don't have to over-invest in hardware that sits idle most of the time; you just
rent the extra capacity from the cloud when you need it, instantly.

3. Managing Massive Data: Beyond Traditional Databases

To handle the enormous scale of the internet, cloud computing relies on new data architectures:
  Distributed File Systems (e.g., GFS/HDFS): Instead of storing a huge file on one disk, these
systems break files into smaller chunks (typically 64 MB) and replicate them (usually three
times) across many cheap, commodity servers. This ensures that if one server fails, the data
is still safe and immediately accessible elsewhere (Fault Tolerance).
 NoSQL Data Stores (e.g., BigTable, Dynamo): These systems replaced traditional relational
databases for massive web applications. They focus on high availability and fast read/writes
rather than strict consistency. Data is stored simply as a key-value map, making it easier to
scale horizontally across thousands of servers.
 MapReduce: This is the clever programming model that makes crunching massive datasets
possible. It works in two steps:
1. Map: Takes a huge job and splits it into thousands of smaller, parallel tasks.
2. Reduce: Collects the results from all the small tasks, groups them, and processes
them into a final answer.
o This allows Google or Facebook to analyze petabytes of data using a vast cluster of
machines simultaneously.

4. OpenStack: The Open-Source Cloud Framework

OpenStack is the popular, open-source software suite that acts as an "operating system" for building
clouds. Its main components handle everything a cloud needs:
OpenStack
Function (What it Does)
Component
Keystone Identity: Manages user accounts, authorization, and security tokens.
Compute: Manages the Virtual Machines (VMs)—it starts, stops, and provisions
Nova
them.
Networking: Manages all network resources like IP addresses, gateways, and
Neutron
connectivity.
Image Repository: Stores and catalogs the base VM disk images (the templates
Glance
for new servers).
Block Storage: Provides permanent storage volumes that act like a local hard
Cinder
drive for your VMs.
Object Storage: Provides highly scalable, accessible-from-anywhere storage for
Swift
general files and backups.
Export to Sheets

Week-4
Abstract Summary: Private Cloud Implementation (OpenStack, Azure, GCP)
I. Private Cloud Implementation: Meghamala @ IIT Kharagpur
The document details the implementation and use of
Meghamala, a private cloud computing platform at IIT Kharagpur, built using the OpenStack
Cloud Computing platform. Meghamala was conceptualized to provide
on-demand computational and storage resources to the institute's research community.
A. Architecture and Services
Meghamala's physical infrastructure includes
Blade servers, SAN Storage, and NAS. The core services offered are:
 VMs4U – Compute Nodes: Allows users to provision a Virtual Machine (VM) on
demand, to be used as a desktop or to run workloads.
 Virtual Machine Configurations ("Flavors"): Three primary configurations are
available, offering escalating resources:
o IITKGP_regular: 2 VCPUs, 4 GB RAM, 45 GB ephemeral storage.
o IITKGP_large: 4 VCPUs, 8 GB RAM, 45 GB ephemeral storage.
o IITKGP_XLarge: 8 VCPUs, 16 GB RAM, 60 GB ephemeral storage.
 Operating Systems: VMs support guest OSs like Ubuntu 14.04, Centos 7, and Fedora
20.
 Storage: Persistent storage is provided upon request, alongside the default
ephemeral storage.
 Specialized Clusters: Meghamala supports dedicated services like MegHadoop (a
Hadoop cluster) and Megha Data (a data storage service).
B. VM Lifecycle and User Access
The document illustrates the end-to-end VM lifecycle, including creation, access, and
termination.
 VM Creation: Users submit a VMs4U Request form. This includes filling out personal
and VM specification details (preferred VM Name, VM Type, OS, number of VMs, and
purpose). The request requires a multi-step approval process: filling the digital form,
printing and signing a hard copy, and submitting the signed copy to the professor-in-
charge.
 Usage Policy: VMs must be used only for academic purposes. The presence of
inappropriate material may lead to the
immediate termination of the VM(s).
 Access: Once created, a VM can be accessed by the user, for instance, through the
X2Go Client.
 OpenStack Dashboard (Horizon): The system admin can monitor usage summaries
(e.g., Active Instances, VCPU-Hours, GB-Hours) and resource limits (e.g., Floating IPs,
Volumes). They can view details of instances, Cinder volumes, Glance images,
Neutron network rules, and Nova hypervisor statistics and flavors.

II. Public Cloud Platform Services: Microsoft Azure and Google Cloud Platform (GCP)
The latter part of the document introduces two major commercial public cloud platforms
and demonstrates application deployment.
A. Microsoft Azure
Azure is a growing collection of integrated cloud services used by developers and IT
professionals to build, deploy, and manage applications through a global network of data
centers.
 Key Features:
o Hybrid Cloud: Connects cloud and on-premises infrastructure with consistent
capabilities.
o Trusted Cloud: Focuses on security, privacy, compliance, and transparency.
o App Innovation: Offers deeply-integrated cloud services for rapid
development and deployment.
o Azure Web Apps: A highly scalable, self-patching web hosting service.
 Python Web App Deployment Process: The deployment involves using the Azure
portal and command-line tools: installing prerequisites (Git, Python) , configuring a
deployment user , creating a
resource group (a logical container) , defining an
App Service Plan (specifying region, size, SKU, and scale count) , creating the
web app , configuring the
Python version , and deploying code using
local Git.
B. Google Cloud Platform (GCP)
GCP is a set of services that enables developers to build, test, and deploy applications on
Google’s reliable infrastructure.
 Value Propositions:
o Leverage Google’s Infrastructure: Built on the same global network and
infrastructure used by Google Search and YouTube, offering redundancy and
innovation.
o Focus on Product: Developers use managed services (like App Engine and
Cloud Datastore) and developer tools to focus on code rather than system
administration.
o Mix and Match: Offers a wide range of services including virtual machines
(Compute Engine), managed platform (App Engine), object storage (Cloud
Storage), NoSQL datastore (Datastore), and MySQL database (Cloud SQL).
o Scale: Applications can automatically scale up to handle demanding
workloads and scale down when traffic subsides. You pay only for what you
use.
 Service Model Choices: GCP offers both Platform-as-a-Service (PaaS) via App Engine
(focus on code, no maintenance) and
Infrastructure-as-a-Service (IaaS) via Compute Engine (raw virtual machines, high flexibility).
 Web Hosting Example: A simple web page can be hosted by creating a Cloud Storage
bucket, configuring the website properties, and making the files publicly accessible.

50 Most Important NPTEL Exam Questions (MCQ)

Section A: OpenStack and Meghamala Implementation
Q1. What is the cloud computing platform utilized to build the Meghamala private cloud at
IIT Kharagpur? A. Microsoft Azure
B. Amazon AWS C. OpenStack
D. Google Cloud Platform Correct Answer: C
Q2. What is the primary purpose of the Meghamala initiative? A. To offer commercial cloud
services globally. B. To provide on-demand computational and storage resources to the
institute research community.
C. To host the institute's official website. D. To sell SAN Storage and Blade servers. Correct
Answer: B
Q3. Which of the following is NOT listed as a physical hardware component of the
Meghamala system? A. Blade servers
B. SAN Storage
C. NAS
D. Desktop PCs Correct Answer: D
Q4. What is the name of the service Meghamala offers for provisioning Virtual Machines?
A. MegData B. VMs4U
C. MegHadoop D. CloudCompute Node Correct Answer: B
Q5. Which virtual machine configuration offered by Meghamala has the highest ephemeral
storage capacity? A. IITKGP_regular (45 GB)
B. IITKGP_large (45 GB)
C. IITKGP_XLarge (60 GB)
D. All have 45 GB Correct Answer: C
Q6. What is the RAM specification for the IITKGP_large VM configuration? A. 4 GB
B. 8 GB
C. 16 GB D. 45 GB Correct Answer: B
Q7. Which of the following is NOT listed as a supported guest operating system for VMs on
Meghamala? A. Ubuntu 14.04
B. Centos 7
C. Fedora 20
D. Windows Server 2016 Correct Answer: D
Q8. The specialized service MegHadoop on Meghamala is a cluster for: A. General VM
provisioning B. Data storage C. A Hadoop cluster
D. Network management Correct Answer: C
Q9. The official usage policy for VMs on Meghamala states they should be used only for: A.
Commercial purposes. B. Personal entertainment. C. Academic purposes.
D. External consulting projects. Correct Answer: C
Q10. What action may lead to the immediate termination of a VM on Meghamala? A.
Using persistent storage. B. Presence of inappropriate material.
C. Using an IITKGP_regular flavor. D. Accessing the VM via X2Go. Correct Answer: B
Q11. The process of requesting a VM on Meghamala requires the user to submit a signed
hard copy to the: A. CIC Engineers. B. Students on the Meghamala team. C. Professor-in-
charge.
D. OpenStack dashboard. Correct Answer: C
Q12. What is the primary software component of OpenStack used for managing compute
resources (VM provisioning)? A. Glance B. Cinder C. Nova
D. Neutron Correct Answer: C
Q13. Which OpenStack component is primarily used to manage persistent storage
volumes? A. Glance B. Cinder
C. Keystone D. Nova Correct Answer: B
Q14. The OpenStack component that provides the image catalog and repository for VM
disk images is: A. Cinder B. Swift C. Glance
D. Neutron Correct Answer: C
Q15. Network access rules (like Security Group rules) in OpenStack are managed by which
component? A. Nova B. Cinder C. Neutron
D. Keystone Correct Answer: C
Q16. The OpenStack dashboard, which provides a graphical interface for users and
administrators, is known as: A. Nova B. Horizon
C. Swift D. Keystone Correct Answer: B
Q17. In the OpenStack dashboard, the admin can monitor Hypervisor usage details for
VCPUs, RAM, and Storage, which are managed by: A. Glance B. Nova
C. Neutron D. Cinder Correct Answer: B
Q18. What OpenStack concept defines the specific hardware configuration (VCPU, RAM,
Disk) for a VM instance? A. Image B. Volume C. Flavor
D. Security Group Correct Answer: C
Q19. Which client application is mentioned for accessing a newly created VM on
Meghamala? A. SSH B. VNC C. RDP D. X2Go Client
Correct Answer: D
Q20. What is the maximum duration (in days) a VM can be requested for, according to the
VMs4U Request form in the document? A. 30 days B. 60 days
C. 90 days D. 365 days Correct Answer: B
Section B: Microsoft Azure and Deployment
Q21. Microsoft Azure is described as a collection of integrated cloud services used to build,
deploy, and manage applications through a: A. Local area network. B. Single monolithic
datacenter. C. Global network of datacenters.
D. Private VLAN only. Correct Answer: C
Q22. What feature of Azure helps connect cloud and on-premises environments
consistently? A. Public Cloud only B. Consistent hybrid cloud capabilities
C. Pure IaaS Model D. Pure PaaS Model Correct Answer: B
Q23. Azure helps protect assets primarily through a rigorous methodology and focus on
security, privacy, compliance, and: A. Open source only. B. Cost optimization. C.
Transparency.
D. Vendor lock-in. Correct Answer: C
Q24. Azure Web Apps is characterized as a web hosting service that is highly scalable and:
A. Manual-patching. B. Self-patching.
C. Hardware-dependent. D. Only supports Microsoft technologies. Correct Answer: B
Q25. What command-line tool is used in the Azure deployment process to configure a
deployment user? A. az compute B.
az webapp deployment user set
C. az resource group create D. az git set Correct Answer: B
Q26. A Resource Group in Azure is defined as a logical container for what components? A.
External databases only. B. Azure resources like web apps, databases, and storage accounts.
C. All cloud provider resources globally. D. Networking devices only. Correct Answer: B
Q27. An App Service Plan in Azure specifies the location, size, and features of the: A. Local
Git repository. B. Web server farm that hosts the app.
C. User's desktop environment. D. Database schema. Correct Answer: B
Q28. Which factor is NOT defined by an App Service Plan? A. Region
B. Instance size
C. Scale count
D. Source Code Language Correct Answer: D
Q29. What deployment method is used in the Python Web App quickstart example in the
document to push from a local repository to a repository in Azure? A. FTP B. GitHub C.
Local Git
D. Visual Studio Team Services Correct Answer: C
Q30. The URL structure used to view a deployed Azure web app instance, as shown in the
document, is typically: A. [Link] B.
[Link]
C. [Link] D. [Link] Correct Answer: B
Section C: Google Cloud Platform (GCP) Services
Q31. Google Cloud Platform (GCP) is built on the same infrastructure used by Google for
which services? A. Search, YouTube, and Gmail.
B. Chrome browser only. C. Android OS only. D. Only Compute Engine VMs. Correct Answer:
A
Q32. The ability of GCP applications to automatically scale up to handle the most
demanding workloads is a key feature related to: A. Managed services.
B. Manual scaling only. C. Fixed pricing models. D. Shared memory architecture. Correct
Answer: A
Q33. Which principle allows GCP customers to only pay for the computing resources they
actually use? A. Redundancy B. Fixed pricing C. Scale-down capabilities
D. Global Network Correct Answer: C
Q34. When a developer wants to focus only on their code without worrying about
patching or maintenance, which PaaS service on GCP should be used? A. Compute Engine
B. Cloud SQL C. App Engine
D. Cloud Storage Correct Answer: C
Q35. Which GCP service provides access to "raw virtual machines" and offers the most
flexibility (IaaS)? A. App Engine B. Cloud SQL C. Compute Engine
D. Cloud Datastore Correct Answer: C
Q36. What is the NoSQL database service offered by Google Cloud Platform? A. Cloud SQL
B. Cloud Storage C. Cloud Datastore
D. BigQuery Correct Answer: C
Q37. Which GCP service provides flexible object storage with global edge caching? A. Cloud
SQL B. Cloud Storage
C. Compute Engine D. Cloud Datastore Correct Answer: B
Q38. What is the necessary configuration file for deploying an application using Google
App Engine (PaaS)? A. [Link] B. [Link] C.
[Link]
D. [Link] Correct Answer: C
Q39. What tool is used to deploy an App Engine application from the command line? A.
dev_appserver.py B.
gcloud app deploy
C. az webapp deploy D. kubectl deploy Correct Answer: B
Q40. What is the initial step for hosting a simple web page inside Google Cloud Platform,
as described in the example? A. Create a Compute Engine VM. B. Create a Cloud Storage
bucket.
C. Install the Flask framework. D. Configure the MySQL database. Correct Answer: B
Section D: Core Cloud Concepts and General Knowledge
Q41. A logical container into which Azure resources like web apps, databases, and storage
accounts are deployed and managed is called a: A. Resource Group
B. App Service Plan C. Virtual Network D. Deployment User Correct Answer: A
Q42. In the App Engine development process, which local server command is used to start
the local development server? A. gcloud app deploy B.
dev_appserver.py
C. gcloud app browse D. [Link] Correct Answer: B
Q43. The storage in Meghamala that is local to the VM and persists only until the VM is
terminated is referred to as: A. Persistent storage
B. SAN Storage C. Ephemeral storage
D. NAS Correct Answer: C
Q44. Which term describes the cloud service model where the user is provided with raw
virtual machines and has the flexibility to build anything they need? A. SaaS (Software as a
Service) B. PaaS (Platform as a Service)
C. IaaS (Infrastructure as a Service)
D. FaaS (Function as a Service) Correct Answer: C
Q45. What is the process that allows a local development server, like the one in App
Engine, to automatically update the running application when source files are changed? A.
Hot-deployment
B. Manual restart C. Containerization D. Continuous Integration
Correct Answer: A (Implicitly by the description: "The development server watches for
changes... and reloads them if necessary" )
Q46. The component in OpenStack that handles identity management, token issuance, and
service catalog is typically: A. Nova B. Neutron C. Keystone
D. Cinder
Correct Answer: C (Inferred from general OpenStack architecture, often covered in Week 3,
and the dashboard login )
Q47. Which term describes the ability of cloud systems to support rapid development,
deployment, and iteration of applications without worrying about underlying system
administration? A. Redundancy B. Managed Services
C. Scale-up only D. Transparent maintenance Correct Answer: B
Q48. The primary advantage of Google Cloud Platform being built on the same
infrastructure as Google Search and YouTube is providing: A. Simple web hosting. B. Global
Network and Redundancy.
C. Free services to all users. D. Only local connectivity. Correct Answer: B
Q49. For the VM Creation request on Meghamala, what is the default number of VMs
listed in the request form? A. 0 B. 1
C. 2 D. Unlimited Correct Answer: B
Q50. The core philosophy of a major public cloud, like Azure or GCP, is that developers get
the freedom to build and deploy wherever they want, using: A. Only proprietary tools. B.
Only Microsoft or Google-approved frameworks. C. The tools, applications, and frameworks
of their choice.
D. Only virtual machines. Correct Answer: C
summary humanier free
This is a more concise, human-friendly summary of the key concepts from the uploaded
content, focusing on private cloud setup and major public cloud offerings.

Human-Friendly Cloud Computing Summary: Building and Using Cloud Services

This document breaks down how modern cloud services are built, using a real-world
example of a university's private cloud and showing how public cloud platforms work.
1. Building a Private Cloud: The Meghamala Example
IIT Kharagpur's Meghamala is a perfect example of a Private Cloud—a system built for a
specific organization's exclusive use.
 The Blueprint: OpenStack: Meghamala is built entirely on OpenStack, an open-
source framework that acts as the operating system for the cloud.
 The Goal: On-Demand Access: Its primary mission is to give researchers on-demand
access to powerful computers (VMs4U service) and storage for academic work.
 The Hardware: It runs on professional-grade gear like Blade Servers and specialized
storage (SAN/NAS).
 The User Journey:
o To get a Virtual Machine (VM), a user must fill out a request form, specify the
required size (Flavor, like IITKGP_large), and get approval from the professor-
in-charge.
o The VM is created with temporary (Ephemeral) storage and can be accessed
via a client like X2Go.
o Strict Rule: VMs must be used only for academic purposes; misuse leads to
immediate termination.
2. OpenStack Components: The Cloud's Core Engine
OpenStack is modular, meaning different parts handle different jobs:
Component Job Description Analogy
Nova The Engine: Manages and provisions all the Virtual Machines (VMs).
Keystone The Gatekeeper: Handles user identity, security, and authorization
Component Job Description Analogy
(login).
Neutron The Router: Manages all networking—IPs, gateways, and connectivity.
The Library: Stores and catalogs all the base operating system images
Glance
(the templates for new VMs).
The Permanent Drive: Manages persistent Block Storage (like a
Cinder
permanent hard disk for the VM).
The File Warehouse: Manages scalable Object Storage for general files
Swift
and backups, accessible from anywhere.
Export to Sheets
3. Public Cloud Giants: Azure & GCP
Major public clouds like Microsoft Azure and Google Cloud Platform (GCP) offer services built
on a massive, global scale.
Feature Microsoft Azure Google Cloud Platform (GCP)
Emphasizes Hybrid Cloud, Built on the same reliable global
Infrastructure seamlessly connecting a user's local network as Google Search and
data center to the public cloud. YouTube.
Focus on Managed Services, allowing
Azure Web Apps for highly scalable,
Key Offering developers to focus on code, not
self-patching web hosting.
maintenance.
App Service Plan defines the Offers PaaS (App Engine, zero
Service
physical resources (size, region) for maintenance) and IaaS (Compute
Models
hosting web applications. Engine, raw VMs).
Uses Resource Groups as logical Provides Cloud Datastore (NoSQL) and
Key Services containers to manage all application Cloud Storage (file/object storage) with
components together. built-in redundancy.
Applications automatically scale up
The High focus on security, privacy, and
and down—you only pay for what you
Advantage compliance (Trusted Cloud).
use.

Week-5
1. Service Level Agreement (SLA)
 Definition: A formal contract between Service Provider (SP) and Service Consumer
(SC) defining performance guarantees.
 Components:
o Service Level Objectives (SLOs) → measurable metrics like uptime, response
time.
o Penalty/Service credits → compensation when SLA is violated.
  Example Problems:
o Availability calculation (% uptime vs outage hours).
o Service cost adjustments with downtime and penalty clauses.

2. Cloudonomics (Economics of Cloud Computing)

 Key Properties:
o Common Infrastructure, Location Independence, Online Connectivity.
o Utility Pricing (pay-per-use).
o On-Demand Elastic Resources.
 Cost Models:
o Baseline cost (owning infrastructure).
o Cloud cost (pay per use).
o Utility Premium (U = Cloud cost ÷ Baseline cost).
o Cloud is cheaper if U<PeakAverageU < \frac{Peak}{Average}U<AveragePeak.
 Penalty Costs: Arise from mismatch between demand D(t)D(t)D(t) and resources
R(t)R(t)R(t).
 Hybrid Model: Mix of owned and rented resources.
 Assignments: Calculation of penalty with delayed provisioning, cost comparison of
in-house vs cloud deployment.

3. MapReduce Programming Model

 Developed by Google for large-scale distributed computation.
 Phases:
o Map: input split into key-value pairs processed in parallel by mappers.
o Shuffle/Sort: intermediate grouping by key.
o Reduce: aggregation of values per key.
 Features: Fault tolerance, scalability, parallelism.
 Hadoop: Open-source implementation.
 Problems discussed:
o File block calculation in HDFS.
o MapReduce pseudo-code for average calculation, salary aggregation, word
categorization.

4. Resource Management – I (Green Cloud Computing)

 Need: Data centers consume ~0.5% of global electricity; energy doubles every 5–6
years.
 Motivation: Economic (high costs), Environmental (CO2 emissions).
 Green Cloud Framework:
o VM scheduling (multi-core power optimization).
o Power-aware & thermal-aware scheduling.
o VM Management (live migration, dynamic shutdown/start).
o Lightweight VM images (reduce boot times, energy savings).
 Outcome: Significant savings in power cost and improved efficiency.

5. Resource Management – II (IaaS Cloud)

 Resources: Physical (CPU, memory, storage, network), Logical (execution,
monitoring).
 Objectives: Scalability, QoS, low latency, cost-effectiveness.
 Challenges: CPU bottlenecks, memory/storage issues, load balancing, security.
 Resource Management Aspects:
 o Provisioning, Allocation, Adaptation, Mapping, Scheduling, Brokering.
 Approaches:
o Provisioning: Game theory, queuing models, adaptive provisioning.
o Allocation: Market-oriented, energy-aware, dynamic scheduling.
o Mapping: Load-aware, minimum congestion, local search, API-based.
o Adaptation: Reinforcement learning, DNS-based load balancing, hybrid
approaches.
 Performance Metrics: Reliability, QoS, Delay, Ease of deployment, Control
overhead.

📘 50 MCQs with Options (Cloud Computing – Week 5)

SLA (Service Level Agreement)

1. SLA is a contract between:
A) Service Provider & Network Operator
B) Service Provider & Service Consumer
C) Consumer & Consumer
D) Provider & Vendor
✅ Answer: B
2. SLA contains measurable conditions called:
A) SLC
B) KPIs
C) SLOs
D) QoS
✅ Answer: C
3. The purpose of SLA is to:
A) Increase profit of SP
B) Define security rules only
C) Define performance and availability guarantees
D) Replace consumer contracts
✅ Answer: C
4. If uptime = 99% per month, allowable downtime (30 days, 24h/day) is:
A) 3.6 hrs
B) 7.2 hrs
C) 12 hrs
D) 15 hrs
✅ Answer: B
5. SLA violation occurs when:
A) Downtime < allowable outage
B) Downtime > allowable outage
C) Uptime = 100%
D) Service credits are not given
✅ Answer: B

Cloudonomics (Economics of Cloud)

6. Cloud pricing model is usually based on:
A) Subscription
B) Utility / Pay-per-use
C) License model
D) Freeware
✅ Answer: B
7. Utility Premium (U) is defined as:
A) Peak demand / Average demand
B) Cloud cost ÷ Baseline cost
C) Average cost ÷ Peak cost
D) Owned cost ÷ Rental cost
✅ Answer: B
8. Cloud is cheaper if:
A) U < Peak/Average demand
B) U > Peak/Average demand
C) U = Average cost
D) Cloud cost = Baseline cost
✅ Answer: A
9. Hybrid cloud cost model means:
A) Only owned resources used
B) Only cloud used
C) Combination of owned + cloud resources
D) Outsourcing resources only
✅ Answer: C
10. Penalty cost in cloudonomics is proportional to:
A) |Peak – Average|
B) |Demand – Resources| over time
C) Number of users
D) Fixed subscription fees
✅ Answer: B

MapReduce
11. MapReduce was originally developed by:
A) Amazon
B) Google
C) Microsoft
D) IBM
✅ Answer: B
12. The open-source implementation of MapReduce is:
A) Spark
B) Hadoop
C) BigTable
D) EC2
✅ Answer: B
13. The two major phases in MapReduce are:
A) Store & Retrieve
B) Shuffle & Load
C) Map & Reduce
D) Send & Receive
✅ Answer: C
14. Map output is stored as:
A) XML files
B) Key-value pairs
C) JSON
D) Tables
✅ Answer: B
15. Reducers aggregate results based on:
A) Mapper ID
B) Keys
C) Values
D) File size
✅ Answer: B
16. Fault tolerance in MapReduce is achieved by:
A) Restarting Hadoop
B) Re-executing failed tasks
C) Using more reducers
D) Replicating results
✅ Answer: B
17. Default block size in HDFS (as given in lecture):
A) 16 MB
B) 32 MB
C) 64 MB
D) 128 MB
✅ Answer: C
18. A file of 127 MB with 64 MB block size creates:
A) 1 block
B) 2 blocks
C) 3 blocks
D) 4 blocks
✅ Answer: C
19. The Word Count program is a classic example of:
A) MapReduce
B) SparkSQL
C) Cloudonomics
D) Virtualization
✅ Answer: A
20. The master process in MapReduce coordinates:
A) Only reducers
B) Only mappers
C) Mappers and reducers
D) Only input storage
✅ Answer: C

Resource Management I (Green Cloud Computing)

21. Data centers consume about what % of global electricity?
A) 0.1%
B) 0.5%
C) 2%
D) 5%
✅ Answer: B
22. Server energy demand doubles every:
A) 2–3 years
B) 4 years
C) 5–6 years
D) 10 years
✅ Answer: C
23. Green Cloud Computing mainly focuses on:
A) Faster booting
B) Energy efficiency
C) More revenue
D) Higher salaries
✅ Answer: B
24. VM scheduling saves energy by:
A) Using GPUs only
B) Optimizing multi-core CPU usage
C) Reducing memory usage
D) Avoiding disk I/O
✅ Answer: B
25. Power-aware scheduling ensures:
A) VMs consume more power
B) Workload is distributed evenly across nodes
C) Maximizing utilization per active core
D) Thermal balancing
✅ Answer: C
26. Thermal-aware scheduling helps to:
A) Reduce overheating and cooling costs
B) Increase CPU clock speed
C) Save network bandwidth
D) Improve latency only
✅ Answer: A
27. VM management involves:
A) Only shutdown of VMs
B) Only load balancing
C) Migration, shutdown/startup, load monitoring
D) Encryption
✅ Answer: C
28. Lightweight VM images help in:
A) Slower boot times
B) Higher energy use
C) Reducing boot time & energy consumption
D) Network congestion
✅ Answer: C
29. A minimal cloud Linux image excludes:
A) Kernel
B) X11/GUI
C) Command line tools
D) File system
✅ Answer: B
30. Boot time reduction from 38s to 8s saves:
A) CPU cycles only
B) Significant energy & money yearly
C) Network latency only
D) RAM usage
✅ Answer: B

Resource Management II (IaaS)

31. IaaS typically provides:
A) Only SaaS
B) VMs, storage, and networks
C) Databases only
D) Software packages only
✅ Answer: B
32. Resource provisioning means:
A) Mapping user tasks to servers
B) Allocating provider’s resources to customers
C) Measuring QoS
D) Creating new APIs
✅ Answer: B
33. Resource allocation refers to:
A) Monitoring resource use
B) Distributing resources economically
C) Shutting down servers
D) Pricing policies
✅ Answer: B
34. Resource mapping means:
A) Creating new hardware
B) Matching demand with supply
C) Encrypting resources
D) Scheduling
✅ Answer: B
35. Resource brokering is:
A) Encrypting requests
B) Negotiating resources via agents
C) Creating databases
D) Monitoring load
✅ Answer: B
36. Nash equilibrium approach is used in:
A) Scheduling
B) Game-theory-based provisioning
C) Energy monitoring
D) Fault tolerance
✅ Answer: B
37. Adaptive provisioning helps in:
A) Automatic bottleneck resolution
B) Static resource allocation
C) Hardware maintenance
D) Increasing latency
✅ Answer: A
38. Market-oriented allocation balances:
A) Provider revenue & customer satisfaction
B) Only provider profits
C) Only energy usage
D) SLA violation
✅ Answer: A
39. Energy-aware allocation is inspired by:
A) Birds
B) Ant colony behavior
C) Neural networks
D) Cloud bursting
✅ Answer: B
40. Reinforcement learning is applied in:
A) Static allocation
B) Dynamic provisioning
C) Security
D) Networking
✅ Answer: B
41. DNS-based load balancing scales applications by:
A) Shutting down VMs
B) Replicating VMs/containers dynamically
C) Increasing storage
D) Encrypting queries
✅ Answer: B
42. Hybrid resource adaptation combines:
A) Bin packing & gradient search
B) Round robin & random allocation
C) Thermal & security scheduling
D) Fault tolerance & encryption
✅ Answer: A
43. Performance metrics for resource management include:
A) QoS, delay, reliability
B) Price only
C) Uptime only
D) CPU cycles only
✅ Answer: A
44. Scalability in IaaS means:
A) Reducing resources
B) Ability to handle growing demand
C) Minimizing costs only
D) Faster boot times
✅ Answer: B
45. Optimal Cloud Resource Provisioning (OCRP) considers:
A) Security issues
B) Demand and price uncertainty
C) Network latency
D) Storage allocation
✅ Answer: B

General
46. Cloud computing is based primarily on:
A) Databases
B) Virtualization
C) Encryption
D) Operating systems
✅ Answer: B
47. Hadoop stores intermediate results in:
A) Master server only
B) Local file system of mappers
C) Reducers only
D) External DB
✅ Answer: B
48. SLA violation leads to:
A) Termination of service
B) Service credits/compensation
C) Extra charges to consumer
D) Higher latency
✅ Answer: B
49. Green cloud combines:
A) Power-aware & Thermal-aware scheduling
B) Encryption & Decryption
C) Map & Reduce
D) SaaS & PaaS
✅ Answer: A
50. Main challenge of resource management in IaaS is:
A) SLA writing
B) Efficient allocation under dynamic demand
C) Reducing network speed
D) VM booting
✅ Answer: B

■ SLA (Service Level Agreement)

• Formal contract: Service Provider ↔ Consumer • Contains measurable objectives (SLOs:
uptime %, response time) • If violated → service credits/discounts • Example: 99% uptime
→ ~7.2 hrs downtime/month
■ Cloudonomics (Economics of Cloud) • Pay-per-use (Utility pricing) • Baseline vs Cloud:

(own + cloud) is best • Penalty cost ∝ |Demand – Resources|

own infra = fixed, cloud = elastic • Cheaper if: U < Peak/Average demand • Hybrid model

■ MapReduce • Developed by Google (Hadoop = open-source) • Steps: Map (split to key–

value), Reduce (aggregate results) • Features: Parallel, scalable, fault-tolerant • Example:
Word count, salary aggregation • HDFS Block size = 64 MB
■ Resource Management I (Green Cloud) • Problem: Data centers use 0.5% of world
electricity, doubling every 5–6 yrs • Green Cloud = energy-efficient design • Techniques:
VM scheduling, Power-aware, Thermal-aware • VM Management: migrate, shutdown idle,
restart on demand • Lightweight VM Images → faster boot, less energy (38s → 8s =
millions saved)
■ Resource Management II (IaaS) • Provides: VMs, Storage, Networking • Goals:
Scalability, QoS, Reliability, Low delay, Cost-effective • Challenges: CPU, memory, storage,
network, energy, security • Aspects: Provisioning, Allocation, Mapping, Adaptation,
Brokering • Approaches: Game theory, Queuing models, Energy-aware (ant colony),
Reinforcement learning, DNS load balancing, Hybrid
■ Quick Key Points • SLA = Promise & penalty • Cloudonomics = Pay-per-use + Hybrid •
MapReduce = Map + Reduce • Green Cloud = Save energy (scheduling + VM management)
• IaaS RM = Provision, Allocate, Map, Adapt, Broker.

Week-7

1. Scheduling in Cloud Computing

 Definition: Process of deciding which job/task runs where and when, on available
resources.
 Goals: Minimize completion time, maximize resource use, reduce cost, ensure
fairness.
 Types of scheduling:
o Static vs Dynamic → Static = pre-decided, Dynamic = adapts at runtime.
o Centralized vs Distributed → Central = one controller, Distributed = multiple
controllers.
o Batch vs Online → Batch = large jobs grouped, Online = immediate
scheduling.

2. Load Balancing
 Why? To avoid overloading some servers while others are idle.
 Benefits: Better utilization, reduced response time, scalability.
 Techniques:
o Round Robin: Equal distribution, no load awareness.
o Weighted Round Robin: Based on server capacity.
o Least Connection: Direct new requests to server with fewest connections.
o Randomized: Assign randomly.
o Dynamic Load Balancing: Based on current workload.

3. Energy-Aware Scheduling
 Data centers consume huge energy → need to reduce power.
 Approaches:
o Consolidation of VMs on fewer servers.
o Switching off idle machines.
o Power-aware scheduling (workload mapped to reduce wattage).
o Thermal-aware scheduling (distributes heat to avoid cooling costs).

4. Workflow Scheduling in Cloud

 Many tasks in cloud are interdependent (scientific workflows, big data jobs).
 Workflow scheduling maps tasks to resources considering deadlines, cost, and
dependencies.
 Algorithms used:
o Greedy: Pick best local option (fast but not always optimal).
o Heuristic: Approximate rules for scheduling.
 o Metaheuristic: Advanced optimization → Genetic Algorithm (GA), Ant
Colony Optimization (ACO), Particle Swarm Optimization (PSO).
 QoS parameters: Execution time, throughput, cost, energy efficiency, reliability.

5. Popular Algorithms
 First-Come First-Serve (FCFS): Simple but unfair if long jobs arrive first.
 Shortest Job First (SJF): Reduces waiting time, but may starve large jobs.
 Min-Min: Assigns tasks with least completion time to fastest resources.
 Max-Min: Assigns longer tasks first to balance load.
 Genetic Algorithm (GA): Uses crossover + mutation for optimization.
 ACO & PSO: Nature-inspired optimization (ants → pheromones, particles → swarm
movement).

✅ In short: Week 7 is all about scheduling strategies in cloud, balancing between

performance, cost, and energy, using both simple (FCFS, Round Robin) and advanced (GA,
ACO, PSO, heuristic) methods.

🎯 50 Most Important MCQs (with Options + Answers)

Scheduling Basics
1. Scheduling in cloud computing is about:
A) Assigning IP addresses
B) Deciding tasks run on which resources
C) Creating SLAs
D) Encrypting jobs
✅B
2. Goal of scheduling is:
A) Increase delay
B) Minimize cost/time and maximize utilization
C) Always reduce energy only
D) Run jobs randomly
✅B
3. Static scheduling means:
A) Fixed at compile time
B) Changes dynamically
C) Based on SLA only
D) Random
✅A
4. Dynamic scheduling adapts:
A) Only before execution
B) At runtime based on workload
C) Once a month
D) Never
✅B
5. Distributed scheduling uses:
A) A single master
B) Multiple controllers
C) Network switches only
D) Random nodes
✅B
Load Balancing
6. Load balancing prevents:
A) SLA creation
B) Overloading some servers while others idle
C) Energy saving
D) Random execution
✅B
7. Round Robin scheduling assigns tasks:
A) Based on power
B) Equally in rotation
C) Only to fastest server
D) Randomly
✅B
8. Weighted Round Robin considers:
A) Server capacity
B) Random number
C) Energy efficiency
D) SLA violations
✅A
9. Least Connection strategy assigns jobs to:
A) Server with smallest load
B) Server with least active connections
C) Server with biggest memory
D) Random machine
✅B
10. Randomized load balancing is:
A) Always optimal
B) Assigning requests randomly
C) Based on SLA
D) Only for workflows
✅B

Energy-Aware Scheduling
11. Data centers consume high:
A) RAM
B) Energy
C) Storage
D) IP addresses
✅B
12. Energy saving can be done by:
A) Overloading servers
B) Consolidating VMs
C) Always shutting down all servers
D) Avoiding virtualization
✅B
13. Thermal-aware scheduling helps in:
A) Better memory usage
B) Balancing heat to reduce cooling cost
C) Increasing latency
 D) Encrypting jobs
✅B
14. Power-aware scheduling is about:
A) Minimizing watt consumption
B) Maximizing number of servers
C) Reducing memory load
D) Scheduling randomly
✅A
15. Green computing goal is:
A) Use more servers
B) Reduce energy with minimal performance impact
C) Ignore SLA
D) Replace CPUs
✅B

Workflow Scheduling
16. Workflow scheduling is important for:
A) Independent tasks only
B) Dependent tasks and scientific jobs
C) Randomized jobs
D) SLA monitoring only
✅B
17. QoS in workflow includes:
A) Execution time, cost, reliability
B) Energy only
C) Encryption only
D) IP address assignment
✅A
18. Greedy algorithms work by:
A) Choosing best local option quickly
B) Searching globally
C) Random pick
D) Energy optimization only
✅A
19. Genetic Algorithm (GA) uses:
A) Queue scheduling
B) Crossover and mutation
C) Thermal balancing
D) Round robin
✅B
20. ACO is inspired by:
A) Bees
B) Ants leaving pheromone trails
C) Birds flying
D) Fish swimming
✅B

Algorithms
21. FCFS drawback:
A) Starves small jobs if long job first
 B) Starves long jobs
C) Ignores SLA
D) Complex to implement
✅A
22. Shortest Job First may:
A) Reduce average waiting time
B) Starve long jobs
C) Both A & B
D) None
✅C
23. Min-Min selects:
A) Task with longest completion time
B) Task with shortest completion time first
C) Random task
D) SLA highest priority
✅B
24. Max-Min selects:
A) Longest task first
B) Shortest task first
C) Random
D) Energy-efficient job
✅A
25. PSO is inspired by:
A) Bees
B) Birds/Fish swarms
C) CPU design
D) Queues
✅B

General Scheduling Concepts

26. Online scheduling means:
A) Scheduling jobs as they arrive
B) Pre-scheduled batch jobs
C) No scheduling
D) Static scheduling
✅A
27. Batch scheduling means:
A) One by one
B) Jobs grouped together for processing
C) Only energy-based
D) Random scheduling
✅B
28. Centralized scheduling has disadvantage of:
A) Simple management
B) Single point of failure
C) Distributed fairness
D) Randomness
✅B
29. Heuristic algorithms are:
A) Exact optimal
 B) Approximate, rule-based
C) Always random
D) Static
✅B
30. Metaheuristic algorithms are:
A) Nature-inspired global optimization methods
B) Static scheduling
C) Random assignments
D) Simple greedy methods
✅A

Advanced Scheduling Topics

31. SLA-aware scheduling considers:
A) Performance only
B) Cost + QoS requirements
C) Random jobs
D) Network delay only
✅B
32. Consolidation in VM scheduling means:
A) Splitting VMs to many servers
B) Packing VMs to fewer active servers
C) Shutting down data centers
D) Load balancing manually
✅B
33. Energy saving is maximized when:
A) All servers run at 100%
B) Idle servers shut down
C) Tasks are delayed indefinitely
D) More VMs created
✅B
34. Reliability in scheduling ensures:
A) Every task meets deadline
B) Every task eventually executed without failure
C) Tasks encrypted
D) Jobs mapped randomly
✅B
35. Hybrid scheduling combines:
A) Static + Dynamic
B) Centralized + Distributed
C) Energy + QoS
D) All of the above
✅D

More MCQs
36. Main reason for scheduling in cloud is:
A) High storage
B) Resource sharing among many users
C) High SLA violations
D) Network bandwidth only
✅B
 37. Round robin disadvantage:
A) Ignores server capacity
B) Too complex
C) Requires GA
D) High energy usage
✅A
38. Workflow mapping difficulty:
A) Dependencies between tasks
B) Too many CPUs
C) Energy always cheap
D) Random costs
✅A
39. Greedy algorithm limitation:
A) Fast but not globally optimal
B) Always global
C) Energy inefficient
D) Cannot run in cloud
✅A
40. Metaheuristics are good for:
A) NP-hard scheduling problems
B) Only simple jobs
C) SLA writing
D) IP allocation
✅A

Final 10 Quick MCQs

41. Energy-aware scheduling reduces: Cost & CO2.
42. Min-Min algorithm focuses on: Shortest jobs first.
43. Max-Min algorithm reduces: Starvation of long tasks.
44. Genetic Algorithm is: Evolution-inspired optimization.
45. ACO scheduling uses: Pheromone-based paths.
46. PSO simulates: Bird flocking/Fish schooling.
47. Online scheduling is needed for: Dynamic arrival of tasks.
48. Thermal-aware scheduling reduces: Cooling cost.
49. SLA-aware scheduling ensures: Agreed QoS is met.
50. Cloud scheduling aims to balance: Performance, Cost, and Energy.

✅ You now have a 2-page summary + 50 MCQs with answers for Week 7 NPTEL exam prep.
Would you like me to also merge Week 5 + Week 7 into a combined PDF (summary +
MCQs) so you have a single exam revision file?
You said:
make summary humanier free
ChatGPT said:
Got it 👍 — I’ll rewrite the Week 7 summary in a human-friendly, simple language (not
heavy academic tone), so it’s easy to grasp and remember for exams.

🌐 Cloud Computing – Week 7 (Human-Friendly Summary)

1. What is Scheduling in Cloud?

Think of scheduling like a traffic cop for tasks. Many users send jobs to the cloud at the
same time. Scheduling decides which task runs on which machine, and when.
 Why needed? To save time, use resources better, reduce costs, and keep things fair.
 Types:
o Static → Pre-planned, fixed before execution.
o Dynamic → Adjusts on the fly, depending on current load.
o Centralized → One main controller.
o Distributed → Many local controllers.
o Batch → Jobs are grouped and then scheduled.
o Online → Immediate scheduling as tasks arrive.

2. Load Balancing
Imagine one waiter in a restaurant serving all tables while others are free — that’s bad
load balancing. The goal is to spread tasks evenly across servers.
 Round Robin → Give jobs one by one in order (simple, but ignores server strength).
 Weighted Round Robin → Assign based on capacity of each server.
 Least Connection → Send new request to the server with the fewest active users.
 Random → Just random distribution (easy, but not always efficient).

3. Energy-Aware Scheduling
Cloud data centers use a lot of electricity. Energy-aware scheduling tries to save power
without harming performance.
 How?
o Combine jobs onto fewer servers and switch off idle ones.
o Place VMs smartly to use less power.
o Spread workload to avoid overheating (less cooling cost).

4. Workflow Scheduling
Some jobs depend on each other (like steps in a recipe). This is called a workflow.
 Scheduling must respect these dependencies while also keeping costs and
deadlines in check.
 QoS (Quality of Service) here means: tasks finish on time, cost is within budget,
results are reliable, and energy is not wasted.
Common Approaches:
 Greedy → Choose the “best now” option (fast but not always optimal).
 Heuristic → Rule-of-thumb solutions, good enough in practice.
 Metaheuristic → Smarter optimization inspired by nature (e.g., Genetic Algorithm,
Ant Colony Optimization, Particle Swarm Optimization).

5. Popular Algorithms
 FCFS (First Come First Serve) → Simple but unfair if a long job comes first.
 SJF (Shortest Job First) → Reduces waiting time but may starve long jobs.
 Min-Min → Schedule shortest tasks first → good for quick wins.
 Max-Min → Schedule longer tasks first → balances workload.
 Genetic Algorithm (GA) → Inspired by evolution, uses “crossover + mutation” to
find better schedules.
 Ant Colony Optimization (ACO) → Mimics ants finding the shortest path with
pheromones.
 Particle Swarm Optimization (PSO) → Inspired by birds/fish moving in swarms to
find the best path.
✅ In short:
 Scheduling = deciding who runs where & when.
 Load Balancing = spread jobs evenly across servers.
 Energy-aware = save electricity & cooling costs.
 Workflow scheduling = map dependent tasks smartly.
 Algorithms = from simple (FCFS, RR) to advanced (GA, ACO, PSO).

Week-8
Docker & Containers
 Docker (2013) = A lightweight way to develop, ship, and run apps anywhere.
 Why useful? → Removes the “works on my machine” problem.
 Features: portable, scalable, lightweight compared to VMs.
 Components:
o Docker Engine → builds & runs containers.
o Docker Hub → online registry of images.
o Docker Compose → manages multi-container apps.
 Containers vs VMs:
o VM → heavy, runs full OS.
o Container → lightweight, shares OS kernel, faster, smaller (MBs).
 Usage: Run anywhere, collaboration, DevOps, faster deployment.

2. Green Cloud Computing

 Definition: Using cloud in an eco-friendly way (reduce CO₂ + energy).
 Why? Data centers consume huge energy & emit as much CO₂ as some countries!
 Energy use in DCs:
o Cooling: ~45%
o IT Equipment: ~40%
o Power Distribution: ~15%
 Solutions:
o Green Brokers → choose cloud services by cost + carbon footprint.
o Green Middleware → software layer optimizing energy use.
o Energy Metrics: PUE (Power Usage Effectiveness).
 CSP Initiatives: Google & Microsoft build data centers near rivers (cheap hydro
power).
 Goal: Balance performance vs energy efficiency.

3. Sensor Cloud Computing

 Motivation: Sensors (RFID, cameras, IoT devices) generate huge data. Alone →
limited. With cloud → scalable.
 Sensor Cloud = Integration of sensor networks + cloud.
 Benefits:
o Collect, share, process, and store massive sensor data.
o Supports real-time processing for smart apps (healthcare, traffic,
environment).
o Offers “Sensing-as-a-Service”.
 Components:
 o Sensor-Cloud Proxy → connects sensors with cloud, manages data.
o Virtual Sensors → software emulation of physical sensors.
 Use Cases: traffic monitoring, smart homes, healthcare (blood sugar, pollution).

4. IoT Cloud
 IoT = Network of connected “things” (sensors, devices, wearables).
 IoT Cloud = IoT + Cloud = scalable data storage, processing, management.
 Architecture:
o Things (devices), Gateways, Cloud.
 Features:
o Scalability (millions of devices).
o Big Data handling.
o Real-time response.
o Highly distributed (edge + cloud).
 Applications: Smart cities, logistics, healthcare, vehicular clouds (intelligent parking,
traffic monitoring).
 Challenge: Security, privacy, data management.

5. Course Summary & Research Areas

 Covered:
o Basics of Cloud, SLAs, Cloud Economics, Security, Resource Management,
Green Cloud, IoT Cloud, Docker, Sensor Cloud.
 Research Areas:
o Fog Computing
o Container Technology
o Hybrid & Federated Clouds
o Cloud Security
o Cloud Analytics (Big Data)

✅ In short:
Week 8 ties together modern technologies (Docker, IoT Cloud, Sensor Cloud) with
sustainability (Green Cloud), and ends with future research areas in cloud computing.

🎯 50 Most Important MCQs (with Answers)

Docker & Containers (1–15)

1. Docker was first released in:
A) 2010 B) 2013 C) 2015 D) 2017
✅B
2. Main motto of Docker is:
A) Encrypt apps B) Develop, ship & run anywhere C) Replace VMs D) Reduce cost
only
✅B
3. Docker containers are:
A) Heavy like VMs B) Lightweight, scalable C) OS kernels D) Hardware-based
✅B
4. Docker Engine is used for:
A) Creating websites B) Building images & running containers C) Managing
databases D) Networking only
✅B
 5. Docker Hub is:
A) Private network B) Image registry C) Data broker D) Security tool
✅B
6. Containers run:
A) On hypervisor B) Directly on host OS kernel C) On cloud only D) In VM only
✅B
7. Which is faster to start?
A) VM B) Container C) Bare-metal OS D) None
✅B
8. Container image contains:
A) Only OS B) OS + App + Dependencies C) App only D) Kernel only
✅B
9. Docker Compose is used for:
A) Writing code B) Managing multi-container apps C) Energy balancing D)
Encryption
✅B
10. VM disadvantage compared to containers:
A) Lightweight B) Needs guest OS per VM C) Portable D) Cloud native
✅B
11. Containers improve:
A) Collaboration B) Isolation C) Scalability D) All of the above
✅D
12. “Works on my machine” problem is solved by:
A) Cloud SLAs B) Docker C) VMs D) Firewalls
✅B
13. A container is:
A) Runtime instance of image B) Static file C) Server D) Host OS
✅A
14. Dockerfile is used for:
A) Logs B) Build script to create images C) Delete files D) Stop containers
✅B
15. Command to list running containers:
A) docker images B) docker ps C) docker run D) docker start
✅B

Green Cloud (16–25)

16. Green computing means:
A) Using green screens B) Eco-friendly computing C) Faster CPUs D) Cloud only
✅B
17. Cloud DCs contribute to:
A) CO₂ emissions B) Rainfall C) Solar energy D) Bandwidth
✅A
18. Largest share of DC energy:
A) IT Equip 40% B) Cooling 45% C) Power dist. 15% D) All
✅B
19. Green Broker considers:
A) Only cost B) Cost + carbon footprint C) Only latency D) Network only
✅B
20. PUE stands for:
A) Power Usage Effectiveness B) Processing Unit Efficiency C) Parallel User
 Execution D) Power Unit Estimation
✅A
21. Amazon estimates energy costs =
A) 10% B) 25% C) 42% D) 60%
✅C
22. Google builds DCs near rivers for:
A) Cooling B) Hydroelectric power C) Cheap land D) Better bandwidth
✅B
23. Idle servers consume ~% of peak load:
A) 0% B) 66% C) 100% D) 20%
✅B
24. Carbon footprint minimization is goal of:
A) Green Cloud B) Sensor Cloud C) IoT Cloud D) Docker
✅A
25. Middleware in Green Cloud helps in:
A) Scheduling B) Energy optimization C) Monitoring D) All
✅D

Sensor Cloud (26–35)

26. Sensor Cloud integrates:
A) Sensors + Cloud B) IoT + Big Data C) Cloud + Green tech D) VMs + Sensors
✅A
27. Sensor-Cloud Proxy manages:
A) Apps only B) Sensor-cloud connectivity & data C) Docker containers D) Databases
✅B
28. Virtual sensor is:
A) Real hardware B) Software emulation of physical sensor C) Container D) Green
broker
✅B
29. Sensor cloud enables:
A) Sensing-as-a-Service B) Mining only C) Only storage D) Cost-only services
✅A
30. Use case of sensor cloud:
A) Smart traffic B) Weather monitoring C) Healthcare D) All of the above
✅D
31. Sensor data lifecycle includes:
A) Collection, processing, storage B) Only storage C) Only encryption D) Only
visualization
✅A
32. Many-to-One virtual config means:
A) One sensor → Many apps B) Many sensors → One aggregated view C) Random
D) Proxy only
✅B
33. Derived virtual sensor config uses:
A) Same type sensors B) Multiple sensor types C) VMs D) Cloud only
✅B
34. Sensor cloud proxy formats data into:
A) PDF B) XML C) HTML D) JSON
✅B
35. Sensor cloud main benefit:
 A) Scalability & sharing sensor data B) Only fast CPUs C) Docker integration D) None
✅A

IoT Cloud (36–45)

36. IoT =
A) Internet of Transactions B) Internet of Things C) Internal OS Tools D) Input-
output Tech
✅B
37. IoT Cloud =
A) IoT devices + Cloud services B) Docker + Cloud C) Green tech only D) Sensor-only
✅A
38. IoT architecture has:
A) Things, Gateway, Cloud B) CPU, RAM, Disk C) VM, Hypervisor, Kernel D) Proxy
only
✅A
39. IoT Cloud advantage:
A) Scalability B) Real-time response C) Big Data handling D) All
✅D
40. IoT Cloud application:
A) Smart city B) Healthcare C) Logistics D) All
✅D
41. IoT Cloud supports:
A) Any time, any place connectivity B) Limited place C) Only offices D) Random
✅A
42. iCOMOT tool manages:
A) IoT Cloud systems B) Docker images C) VMs D) Security keys
✅A
43. IoT challenge:
A) Security & privacy B) Bandwidth only C) Docker only D) None
✅A
44. Vehicular data cloud uses:
A) Naïve Bayes, Logistic Regression B) GA only C) Docker engine D) Sensor proxy
✅A
45. Intelligent Parking is example of:
A) IoT Cloud service B) Green Cloud service C) Docker Hub D) Sensor-only system
✅A

Course & Research (46–50)

46. Fog computing is:
A) Cloud at edge B) Container C) IoT sensor D) Virtualization
✅A
47. Cloud security involves:
A) Privacy, access control, identity mgmt B) Docker images C) Load balancing only
D) VMs only
✅A
48. Microservices architecture improves:
A) Scalability & reliability B) Docker Hub C) VM isolation D) Green Cloud
✅A
49. Big Data + Cloud used for:
A) Analytics B) Monitoring only C) Security only D) Networking
 ✅A
50. Research trend in Cloud:
A) Fog, Sensor Cloud, Container Tech B) Green energy only C) IoT only D) None
✅A

✅ Now you have a 2-page summary + 50 MCQs with answers for Week 8 NPTEL exam prep.
Would you like me to also combine Week 5, 7, and 8 summaries + MCQs into a single
“Exam Master PDF” for full revision?
You said:
make humanier free
ChatGPT said:
Got it 👍 — I’ll make the Week 8 summary more human-friendly, so it reads like simple
revision notes instead of a textbook.

🌐 Cloud Computing – Week 8 (Human-Friendly Summary)

1. Docker & Containers 🐳

 Docker is like a magic box that lets you build once, run anywhere.
 Released in 2013, it solved the “works on my machine” problem.
 Containers are like lightweight VMs:
o Share the same OS kernel → faster, smaller (in MBs).
o Start in seconds (VMs take minutes).
 Key parts:
o Docker Engine → builds and runs containers.
o Docker Hub → online store of container images.
o Docker Compose → runs multiple containers together.
 Why use it? → Portability, easy teamwork, fast deployment, DevOps-friendly.

2. Green Cloud Computing 🌱

 Data centers eat a lot of power → cooling alone takes ~45%.
 This also means a big carbon footprint.
 Green Cloud = run clouds in eco-friendly ways.
 How?
o Shut down idle servers.
o Pack VMs on fewer machines to save energy.
o Place data centers near rivers for hydro power (like Google).
 Metrics: PUE (Power Usage Effectiveness) = Total energy ÷ IT energy.
 Tools: Green brokers (choose service based on cost + energy), middleware
(optimize energy use).

3. Sensor Cloud 📡
 Sensors (RFID, cameras, IoT devices) collect tons of data. Alone → limited.
 Sensor Cloud = put all this sensor data into the cloud.
 Benefits:
o Scale up easily.
o Share sensor data with many apps.
o “Sensing-as-a-Service” → pay and use sensor data like utilities.
 Key ideas:
o Proxy → middleman that connects sensors to the cloud.
o Virtual sensors → software versions of real sensors, can combine many into
 one.
 Uses: Smart traffic, smart homes, pollution monitoring, healthcare.

4. IoT Cloud 🌍
 IoT = Internet of Things → smart devices connected everywhere.
 IoT Cloud = IoT + Cloud → store and process IoT data at scale.
 How it works: Devices → Gateways → Cloud.
 Features:
o Real-time response.
o Can handle millions of devices.
o Works well with Big Data and analytics.
 Examples:
o Smart cities (traffic lights, parking).
o Smart healthcare (wearables, patient monitoring).
o Logistics & transport (fleet tracking).
 Challenges: Security, privacy, and handling so much data.

5. Course Wrap-Up & Research 🚀

 We covered:
o Basics (SLA, Cloud Economics).
o Resource management & Green Cloud.
o Advanced topics → Docker, Sensor Cloud, IoT Cloud.
 Future research areas:
o Fog/Edge Computing → cloud closer to users.
o Containers & microservices → faster and lighter than VMs.
o Hybrid clouds & federated clouds.
o Cloud analytics (Big Data + AI).
o Stronger security & privacy.

✅ Quick Takeaways
 Docker = Fast, portable apps.
 Green Cloud = Save energy + reduce CO₂.
 Sensor Cloud = Share sensor data via cloud.
 IoT Cloud = Smart devices powered by cloud.
 Research = Edge, containers, analytics, and security.

Week-9
1. Cloud–Fog–Edge Computing Overview
Cloud computing provides on-demand services (“Anything-as-a-Service”) from centralized
data centers. However, as IoT devices grow, massive data generation causes latency,
bandwidth, and scalability issues in pure cloud models.
To solve this, Fog Computing extends cloud capabilities to the network edge—closer to
users—allowing real-time data processing. Coined by Cisco, fog computing supports
billions of connected IoT devices by distributing computing, storage, and control tasks.
Edge Computing lies even closer to devices, where sensors and gateways handle
lightweight computations.
Together, Cloud–Fog–Edge Computing forms a layered architecture that brings intelligence
“down from the cloud” to the ground, reducing latency and network load while improving
scalability and cost-efficiency.

2. Cloud vs Fog vs Edge

Aspect Cloud Fog Edge
Centralized Data Intermediate layer between Device-level (Routers,
Location
Centers Cloud & Edge Sensors)
Latency High Moderate Low
Processing Remote Near user On device
Example AWS, Azure Local Gateway Smart sensors
Fog benefits include real-time responses, reduced congestion, cost savings, and better
location awareness.

3. Health Cloud–Fog Case Study

A multi-layered IoT health monitoring system demonstrates fog computing’s efficiency:
 Cloud layer: Long-term storage & analytics
 Fog layer: Real-time patient data analysis (Area Gateway)
 Edge layer: Wearable sensors & mobile devices
Results from simulations using iFogSim:
 Network usage: Lower in fog as only critical data goes to cloud.
 Latency: Fog architecture yields constant, low latency.
 Execution cost: Lower, as cloud resource use is minimized.
 Energy consumption: Reduced by processing near data sources.
Prototype implementation used Raspberry Pi as fog nodes and AWS as the cloud.

4. Resource Management in Cloud–Fog Systems

Processing IoT data solely in the cloud is inefficient for time-sensitive applications.
Fog and Edge computing help manage workloads closer to data sources.
Challenges: Resource allocation, workload balancing, provisioning, scheduling, and
maintaining QoS.
Goal: Efficient resource utilization and reduced latency.
Architecture Layers:
1. Client/Edge Layer – User devices, sensors
2. Fog Layer – Local computation
3. Cloud Layer – Centralized processing and analytics

5. Resource Management Approaches

 Architecture-based:
o Data Flow – movement of workloads (Cloud ↔ Edge).
o Control – centralized vs distributed management.
o Tenancy – single vs multi-tenant hosting.
 Infrastructure-based:
 o Hardware – Gateways, routers, Wi-Fi APs, micro data centers.
o System Software – OS and virtualization.
o Middleware – Handles coordination, orchestration, and deployment.
 Algorithm-based:
o Discovery – Identify available edge resources.
o Benchmarking – Evaluate performance metrics.
o Load Balancing – Efficient workload distribution.
o Placement – Optimal task placement on available nodes.

6. Service Placement Problem

Refers to mapping application components to suitable nodes across Cloud–Fog–Edge
layers while meeting:
 Resource constraints (CPU, RAM, bandwidth)
 Network constraints (latency, bandwidth)
 Application constraints (locality, deadlines)
Optimization objectives: minimize latency, cost, energy, and maximize utilization.

7. Offloading Techniques
Offloading involves shifting computation or data:
 User Device → Edge: Application partitioning and caching.
 Cloud → Edge: Move workloads closer to data origin for efficiency.
Improves real-time response and reduces bandwidth usage.

8. Hardware, Software, and Middleware

 Hardware: Low-power fog devices (routers, gateways, IoT nodes).
 System Software: Virtualization & network management for multi-tenancy.
 Middleware: Provides APIs, communication, coordination, and orchestration
between fog layers.
Algorithms include discovery, benchmarking, load balancing, and placement.

9. Cloud Federation
A federated cloud is a collaborative environment where multiple Cloud Service Providers
(CSPs) share resources to improve utilization and availability.
Goals: interoperability, capacity utilization, and load balancing.
Types of Federation Architectures:
1. Hybrid/Bursting: Private cloud uses public cloud for peak loads (loosely coupled).
2. Broker: Mediator allocates resources across clouds based on cost/performance.
3. Aggregated: Partner clouds share resources (partially coupled).
4. Multitier: Centralized control of multiple clouds (tightly coupled).

Key Takeaways
 Fog computing reduces latency, cost, and energy consumption.
 Resource management ensures QoS through optimal task placement.
 Service placement and offloading improve efficiency and scalability.
 Cloud federation enables collaboration and interoperability among providers.

🧠 50 Important MCQs for NPTEL Cloud Computing (Week 9)

1. Who introduced the term “Fog Computing”?
A) IBM B) Google C) Cisco D) Microsoft
✅ Ans: C) Cisco
2. Fog computing primarily addresses which issue in cloud computing?
A) Security B) Latency C) Cost D) Backup
✅ Ans: B) Latency
3. Which layer is closest to the IoT devices?
A) Cloud B) Fog C) Edge D) Middleware
✅ Ans: C) Edge
4. Main advantage of fog computing is:
A) Higher latency B) Low latency C) Centralized control D) Expensive
✅ Ans: B) Low latency
5. Fog layer acts as:
A) End-user B) Intermediate layer C) Storage only D) None
✅ Ans: B) Intermediate layer
6. Which tool is used for Fog simulation?
A) iCloud B) iFogSim C) CloudSim D) FogTest
✅ Ans: B) iFogSim
7. Health Cloud–Fog system shows minimum latency when modules are placed in:
A) Cloud B) Fog C) Edge D) ISP
✅ Ans: B) Fog
8. Which is NOT an advantage of Fog computing?
A) Real-time processing B) Reduced bandwidth C) High latency D) Local data
awareness
✅ Ans: C) High latency
9. Resource management includes:
A) Scheduling B) Provisioning C) Allocation D) All
✅ Ans: D) All
10. Main goal of Fog resource management:
A) Increase cost B) Optimize latency and resource usage C) Reduce nodes D)
None
✅ Ans: B
11. Control in Fog systems can be:
A) Centralized B) Distributed C) Both D) None
✅ Ans: C
12. Service placement problem deals with:
A) Scheduling packets B) Mapping apps to resources C) User management D)
Pricing
✅ Ans: B
13. Main optimization metrics in service placement:
A) Latency, cost, energy B) Only cost C) Power only D) None
✅ Ans: A
14. Offloading means:
A) Transferring load B) Deleting files C) Uploading D) Compression
✅ Ans: A
15. Offloading from user to edge improves:
A) Response time B) Cost C) Bandwidth usage D) Both A & C
✅ Ans: D
16. Fog nodes typically include:
A) Data centers B) Routers/Gateways C) Satellites D) None
✅ Ans: B
17. Middleware in Fog provides:
A) Hardware interface B) Communication & orchestration C) Encryption D)
None
✅ Ans: B
18. System software in Fog supports:
A) Multi-tenancy B) Virtualization C) Resource management D) All
✅ Ans: D
19. Which architecture uses public cloud during peak load?
A) Broker B) Hybrid/Bursting C) Aggregated D) Multitier
✅ Ans: B
20. Federation broker mainly helps in:
A) Scheduling B) Load balancing C) Resource allocation among clouds D) All
✅ Ans: D
21. Aggregated architecture is generally:
A) Loosely coupled B) Partially coupled C) Tightly coupled D) None
✅ Ans: B
22. Multitier federation is:
A) Loosely coupled B) Tightly coupled C) Distributed D) None
✅ Ans: B
23. Example of Edge device:
A) Cloud data center B) IoT sensor C) AWS EC2 D) None
✅ Ans: B
24. Which factor reduces energy consumption?
A) Processing near source B) Centralized computing C) High latency D) None
✅ Ans: A
25. Main aim of Cloud Federation:
A) Competition B) Cooperation C) Redundancy D) Isolation
✅ Ans: B

26. What is the main function of the Fog layer in IoT systems?
A) Long-term storage B) Real-time processing C) Backup D) Encryption
✅ Ans: B) Real-time processing
💡 Fog nodes process data closer to devices to ensure real-time responses.

27. Which issue is best addressed by Fog computing?

A) High storage B) Low security C) Latency and bandwidth D) Fault tolerance
✅ Ans: C) Latency and bandwidth

28. Which of the following is a resource constraint in application placement?

A) CPU & Memory B) Cost only C) Energy D) None
✅ Ans: A) CPU & Memory

29. Which of these is NOT a function of the middleware in Fog computing?

A) Orchestration B) Communication C) Virtualization D) Monitoring
✅ Ans: C) Virtualization
💡 Virtualization is handled by system software, not middleware.

30. The control mechanism in resource management can be:

A) Only centralized B) Only distributed C) Both centralized and distributed D)
Manual
 ✅ Ans: C

31. Which of the following is a layer in the Cloud–Fog–Edge architecture?

A) ISP Layer B) Client, Fog, Cloud C) Network Only D) Edge Only
✅ Ans: B

32. The iFogSim toolkit is used to:

A) Simulate machine learning B) Model resource management in Fog C) Store
cloud data D) Manage servers
✅ Ans: B

33. In Fog architecture, network congestion is:

A) Increased B) Decreased C) Unaffected D) Doubled
✅ Ans: B

34. Which parameter is minimized in Fog computing compared to Cloud?

A) Latency B) Cost C) Energy D) All of these
✅ Ans: D

35. Which of the following is an algorithmic approach in Fog resource management?

A) Discovery B) Placement C) Load balancing D) All
✅ Ans: D

36. The term ‘Service Placement Problem’ refers to:

A) Mapping services to suitable nodes B) Placing ads C) Allocating users D)
None
✅ Ans: A

37. Locality constraint ensures that:

A) Data stays within specific regions B) Faster computation C) Random
allocation D) None
✅ Ans: A

38. Energy consumption in Cloud-only architecture is usually:

A) Lower B) Higher C) Same D) Negligible
✅ Ans: B

39. Which metric is used in performance analysis of Fog architectures?

A) Latency B) Energy C) Network usage D) All
✅ Ans: D

40. Offloading from Cloud to Edge primarily improves:

A) Storage B) Speed and latency C) Security D) Backup
✅ Ans: B

41. Fog nodes are generally located at:

A) Cloud Data Centers B) Network edge (Gateways, Routers) C) End-user
devices D) None
✅ Ans: B
 42. Which of the following is NOT part of the resource management architecture?
A) Data flow B) Control C) Tenancy D) Encryption
✅ Ans: D

43. In Cloud–Fog–Edge hierarchy, computation load is mostly shifted to:

A) Central Cloud B) Edge Devices C) Fog Layer D) None
✅ Ans: C

44. System virtualization helps in:

A) Resource duplication B) Multi-tenancy and isolation C) Security
management D) None
✅ Ans: B

45. Network virtualization is a part of which layer?

A) Middleware B) Hardware C) System software D) User interface
✅ Ans: C

46. Which architecture allows partial control and resource sharing between clouds?
A) Hybrid B) Broker C) Aggregated D) Multitier
✅ Ans: C

47. The Hybrid (Bursting) architecture is mainly:

A) Tightly coupled B) Loosely coupled C) Partially coupled D) Fully integrated
✅ Ans: B

48. Broker architecture focuses on:

A) Virtualization B) Load balancing and cost optimization C) Data replication
D) None
✅ Ans: B

49. Which type of federation gives full control over remote resources?
A) Loosely coupled B) Partially coupled C) Tightly coupled D) None
✅ Ans: C

50. Main benefit of Cloud Federation is:

A) Vendor lock-in B) Resource sharing and interoperability C) High latency D)
Less utilization
✅ Ans: B

Cloud–Fog–Edge Computing (Human-Friendly Summary)

1. What is Cloud–Fog–Edge Computing?
Cloud computing means using the internet to access computing services like servers,
storage, and databases (“Anything-as-a-Service”).
It works great for large-scale data processing but has some limits — especially when IoT
(Internet of Things) devices generate huge amounts of data every second.
When millions of sensors and smart devices send data to the cloud, it causes:
 High latency (delay in response),
 Heavy bandwidth usage, and
 Slower real-time performance.
To fix this, a new idea called Fog computing was introduced by Cisco.
Fog computing brings the power of the cloud closer to where the data is created — at the
network edge.
Then comes Edge computing, which takes it one step further: even the devices like routers,
gateways, and sensors can process small amounts of data themselves.
So, the structure looks like this:
IoT Devices → Edge → Fog → Cloud
Each layer processes part of the data, helping systems respond faster and use less network
bandwidth.

💡 2. Why Do We Need Fog Computing?

 The cloud is too far away for real-time responses.
 The edge alone doesn’t have enough resources.
 So, fog computing sits in between — close enough to devices but powerful enough
to handle heavy work.
Main benefits:
 Low latency (quick response)
 Less bandwidth usage
 Lower cost
 Local data awareness
 Real-time analytics

❤️3. Case Study – Health Cloud–Fog System

Imagine a health monitoring system:
 Sensors on a patient’s body collect data (heart rate, temperature, etc.)
 The fog layer at a nearby gateway quickly analyzes it.
 Only important or abnormal data is sent to the cloud for deeper analysis.
This system was tested using a simulator called iFogSim.
Results showed:
 Lower latency and energy use
 Less network load because only critical data went to the cloud
 Lower cost since fog devices handled most processing locally
In real-life experiments, Raspberry Pi devices were used as fog nodes and AWS Cloud acted
as the cloud server.

⚙️4. Resource Management in Cloud–Fog Systems

When lots of IoT devices send data, we need to manage how and where tasks run.
This is called resource management.
It includes:
 Allocating tasks to devices,
 Balancing workloads,
 Scheduling jobs,
 Managing resources (CPU, RAM, bandwidth), and
 Maintaining QoS (Quality of Service).
Fog and Edge computing help by handling tasks near the data source, instead of sending
everything to the cloud.
Layers:
1. Client/Edge layer: sensors, smartphones, user devices.
2. Fog layer: local gateways or micro data centers.
3. Cloud layer: big servers for storage and analytics.
🧩 5. Resource Management Approaches
1. Architecture-based:
o Data Flow: how data moves between layers.
o Control: centralized (one controller) or distributed (many controllers).
o Tenancy: how many apps share a node.
2. Infrastructure-based:
o Hardware: routers, Wi-Fi access points, small servers.
o System software: operating systems, virtualization software.
o Middleware: coordinates communication and task deployment.
3. Algorithm-based:
o Discovery: finding free devices or resources.
o Benchmarking: testing performance.
o Load balancing: distributing work evenly.
o Placement: deciding where each app should run.

🔄 6. Service Placement and Offloading

Service placement means finding the best location (Cloud, Fog, or Edge) to run an
application so that it meets:
 Resource limits (CPU, RAM),
 Network limits (latency, bandwidth),
 Application limits (location, timing).
Offloading means moving part of the workload from one layer to another:
 From user device → edge for faster response.
 From cloud → edge to reduce network traffic.
This helps balance performance and cost.

💻 7. Components of Fog Environment

 Hardware: small, low-power devices (routers, gateways, IoT nodes).
 System Software: manages resources, supports virtualization.
 Middleware: handles communication, coordination, and performance monitoring.

☁️8. Cloud Federation

Cloud Federation means multiple Cloud Service Providers (CSPs) working together and
sharing their resources.
This helps in better load balancing, global reach, and reliability.
Why it’s useful:
 Avoid service interruption,
 Improve interoperability,
 Expand coverage and capacity.
Types of Federation:
1. Hybrid/Bursting: Private cloud uses public cloud during high demand (loose
connection).
2. Broker: A broker chooses the best cloud for each task (based on cost, speed, etc.).
3. Aggregated: Partner clouds share resources (partial connection).
4. Multitier: A central system controls multiple cloud sites (tight connection).

📚 In Short
 Cloud computing handles large-scale, centralized processing.
 Fog computing extends the cloud closer to users for faster response.
  Edge computing enables processing directly on devices.
 Together, they make systems smarter, faster, and more efficient.
 Proper resource management and service placement are key to maintaining
performance and minimizing cost.
 Cloud Federation connects multiple providers for seamless, large-scale operations.

Week -10
Cloud Security and Privacy Paradigm
1. Introduction
Cloud computing provides virtualized, on-demand resources over the Internet, but its
distributed and multi-tenant nature introduces complex security and privacy challenges.
Week 10 of the course focuses on the architectural security framework of clouds, key
vulnerabilities, and corresponding countermeasures.
2. Cloud Security Fundamentals
Security in a cloud environment spans three service layers:
 IaaS: Protecting virtual machines, storage, and network infrastructure.
 PaaS: Securing middleware, APIs, and runtime environments.
 SaaS: Managing application-level threats and data integrity.
Core security requirements include confidentiality, integrity, availability, authentication,
authorization, and non-repudiation.
3. Threat Landscape
Common threats involve:
 Data breaches and leakage caused by weak access control or multi-tenancy.
 Account hijacking through credential theft.
 Insecure interfaces and APIs.
 Denial-of-Service (DoS) attacks targeting availability.
 Malicious insiders misusing administrative privileges.
4. Security Architecture
A multi-layered approach integrates:
1. Identity and Access Management (IAM) for authentication and role-based control.
2. Virtualization Security, addressing hypervisor attacks and VM isolation.
3. Network Security, employing firewalls, IDS/IPS, VPNs, and encryption.
4. Data Security, ensuring encryption both at rest and in transit, supported by key-
management frameworks.
5. Security Monitoring and Auditing, establishing trust and compliance.
5. Privacy Preservation
Data in the cloud may cross jurisdictions, raising legal and ethical issues. Techniques such
as anonymization, homomorphic encryption, and differential privacy are used to protect
user identity and data confidentiality without compromising functionality.
6. Trust Management
Trust is established through Service-Level Agreements (SLAs) and certification (e.g., ISO
27017, CSA STAR). Continuous monitoring and reputation systems reinforce user
confidence.
7. Case Studies
 Amazon AWS Security Model implements shared-responsibility between provider
and user.
 Google Cloud Platform integrates encryption by default and distributed denial-of-
service protection.
8. Research Directions
Emerging concerns include:
 Security in multi-cloud and federated environments.
 Use of blockchain for access auditing.
 Integration of AI-driven threat detection.
9. Conclusion
Securing the cloud requires coordinated mechanisms across technological, organizational,
and legal domains. Effective design of IAM, encryption, auditing, and compliance ensures
reliability, trust, and privacy in large-scale distributed infrastructures.

☁️Section 2 – Human-Friendly Summary

(for quick understanding and exam revision)
Cloud Security & Privacy Made Easy
Cloud computing lets us store and run everything online, but because data lives on shared
servers, we must keep it safe and private.
🔐 Why Security Matters
When many users share the same cloud, hackers or mistakes can lead to:
 Stolen passwords and accounts,
 Data leaks,
 Service downtime (DoS attacks),
 Or even insider misuse.
🧩 How Security Works in the Cloud
Clouds have three layers—IaaS, PaaS, SaaS—and each needs protection.
Layer Example What to Secure
IaaS VMs, Storage Firewalls, VM isolation
PaaS Middleware, APIs Patch management, secure APIs
SaaS Apps, user data Encryption, access control
Basic rules:
 Keep data encrypted (while storing & sending).
 Use strong logins and multi-factor authentication.
 Give permissions only to who really needs them.
🧠 Privacy Protection
Clouds store personal data across countries, so companies use:
 Anonymization – removing personal info,
 Homomorphic encryption – processing data without decrypting,
 Differential privacy – adding noise to prevent identity leaks.
🧱 Building Trust
Cloud providers and users share responsibility. Providers secure hardware; users must
protect their own data and passwords.
Trust also comes from SLAs and certifications (like ISO 27001).
⚙️Real Examples
 AWS: strong identity control + encryption + auditing.
  Google Cloud: default encryption and DDoS defenses.
🚀 Future Trends
 AI-based automatic threat detection,
 Blockchain logs for transparency,
 Cross-cloud security standards.
In short: keep data safe, encrypt everything, use verified clouds, and never share keys
carelessly.

1. Which layer in cloud computing provides virtual machines and storage?

A) SaaS
B) IaaS
C) PaaS
D) DBaaS
✅ Answer: B) IaaS

2. Main security objectives in cloud computing are —

A) Confidentiality, Integrity, Availability
B) Performance, Cost, Speed
C) Scalability, Virtualization, Redundancy
D) Auditability, Replication, Throughput
✅ Answer: A) Confidentiality, Integrity, Availability

3. Which company introduced the shared-responsibility model?

A) IBM
B) Amazon Web Services (AWS)
C) Google Cloud
D) Oracle Cloud
✅ Answer: B) Amazon Web Services (AWS)

4. Data at rest should be protected using —

A) Compression
B) Encryption
C) Virtualization
D) Mirroring
✅ Answer: B) Encryption

5. A Denial-of-Service (DoS) attack primarily affects —

A) Confidentiality
B) Availability
C) Integrity
D) Authentication
✅ Answer: B) Availability

6. The weakest link in multi-tenant cloud environments is —

A) Virtualization
B) API Gateway
C) User Access Control
D) Network Layer
✅ Answer: C) User Access Control
7. Identity and Access Management (IAM) focuses on —
A) Scheduling
B) Authentication and Authorization
C) Encryption
D) Backups
✅ Answer: B) Authentication and Authorization

8. Which of the following is NOT a cloud security threat?

A) Data leakage
B) Account hijacking
C) High availability
D) Insecure API
✅ Answer: C) High availability

9. Homomorphic encryption allows —

A) Data to be decrypted before processing
B) Data to be processed while encrypted
C) Random key generation
D) Anonymous file deletion
✅ Answer: B) Data to be processed while encrypted

10. Differential privacy is used to —

A) Improve system speed
B) Hide individual data information
C) Reduce energy consumption
D) Compress large files
✅ Answer: B) Hide individual data information

11. Malicious insiders are dangerous because —

A) They bypass external defenses
B) They slow down services
C) They encrypt data automatically
D) They manage costs
✅ Answer: A) They bypass external defenses

12. Multi-factor authentication improves —

A) Cost efficiency
B) Security strength
C) Network latency
D) Data compression
✅ Answer: B) Security strength

13. The hypervisor operates at which level?

A) Hardware Layer
B) Virtualization Layer
C) Application Layer
D) API Layer
✅ Answer: B) Virtualization Layer
14. The ISO standard for cloud security best practices is —
A) ISO 9001
B) ISO 27017
C) ISO 42001
D) ISO 31000
✅ Answer: B) ISO 27017

15. A “VM escape” attack targets —

A) Middleware
B) Hypervisor
C) Operating System
D) Database
✅ Answer: B) Hypervisor

16. Most API-related threats occur in —

A) IaaS
B) PaaS
C) SaaS
D) All layers
✅ Answer: B) PaaS

17. Data integrity in cloud means —

A) Data is backed up
B) Data is accurate and unaltered
C) Data is compressed
D) Data is deleted securely
✅ Answer: B) Data is accurate and unaltered

18. The shared-responsibility model means —

A) Cloud provider handles all security
B) Users handle all security
C) Both provider and user share duties
D) No one handles security
✅ Answer: C) Both provider and user share duties

19. Which protocol ensures secure communication in transit?

A) HTTP
B) TCP
C) HTTPS/TLS
D) UDP
✅ Answer: C) HTTPS/TLS

20. Key management is essential for —

A) File compression
B) Encryption and decryption
C) VM migration
D) Backup recovery
✅ Answer: B) Encryption and decryption

21. Cloud auditing ensures —

A) Performance improvement
B) Compliance and accountability
C) Fast deployment
D) Data compression
✅ Answer: B) Compliance and accountability

22. Which attack makes resources unavailable to users?

A) SQL Injection
B) DoS (Denial of Service)
C) Phishing
D) Eavesdropping
✅ Answer: B) DoS (Denial of Service)

23. Virtual Machine isolation prevents —

A) Inter-VM data leakage
B) Latency issues
C) Energy consumption
D) Hardware redundancy
✅ Answer: A) Inter-VM data leakage

24. VPN and firewalls are examples of —

A) Network Security
B) Application Security
C) Data Security
D) System Security
✅ Answer: A) Network Security

25. SLAs (Service Level Agreements) define —

A) Security responsibilities and QoS
B) Server location
C) Storage limit
D) Pricing
✅ Answer: A) Security responsibilities and QoS

26. Encryption at rest protects data when —

A) It is transmitted
B) It is stored
C) It is processed
D) It is deleted
✅ Answer: B) It is stored

27. Cloud privacy mechanisms mainly prevent —

A) Data disclosure
B) System updates
C) Cost increase
D) Backup loss
✅ Answer: A) Data disclosure

28. Which model gives the user maximum control over security?
A) SaaS
B) PaaS
C) IaaS
D) DBaaS
✅ Answer: C) IaaS

29. VPN provides —

A) Public access
B) Encrypted private communication
C) Cloud storage
D) Identity proof
✅ Answer: B) Encrypted private communication

30. IDS (Intrusion Detection System) is used to —

A) Detect malicious activity
B) Encrypt data
C) Compress logs
D) Manage bandwidth
✅ Answer: A) Detect malicious activity

31. CSA STAR certification provides —

A) Cloud Security Assurance
B) Hardware validation
C) Backup testing
D) Application patching
✅ Answer: A) Cloud Security Assurance

32. Cloud forensics focuses on —

A) Legal and incident investigation
B) Resource scheduling
C) Application debugging
D) Performance testing
✅ Answer: A) Legal and incident investigation

33. The major legal issue in cloud storage is —

A) Jurisdiction
B) Throughput
C) Latency
D) Cost
✅ Answer: A) Jurisdiction

34. Intercepting communication between two parties is —

A) Man-in-the-Middle attack
B) Phishing attack
C) SQL Injection
D) DoS attack
✅ Answer: A) Man-in-the-Middle attack

35. Best method to ensure data confidentiality is —

A) Replication
B) Encryption
C) Compression
D) Hashing
✅ Answer: B) Encryption

36. Integrity verification commonly uses —

A) Hash functions (SHA, MD5)
B) VPN
C) Ping command
D) Logs
✅ Answer: A) Hash functions (SHA, MD5)

37. The principle of least privilege means —

A) Granting all permissions
B) Giving only necessary access
C) Denying all access
D) Rotating passwords frequently
✅ Answer: B) Giving only necessary access

38. A major hypervisor vulnerability is —

A) VM escape
B) DoS
C) Phishing
D) Sniffing
✅ Answer: A) VM escape

39. A classic example of an availability attack is —

A) DoS
B) Spoofing
C) Phishing
D) Sniffing
✅ Answer: A) DoS

40. A recommended cloud backup strategy is the —

A) 3-2-1 rule
B) 1-1-1 rule
C) 5-5-5 rule
D) 2-2-2 rule
✅ Answer: A) 3-2-1 rule

41. Blockchain can enhance cloud security by —

A) Improving data integrity and auditability
B) Reducing encryption time
C) Increasing speed
D) Decreasing cost only
✅ Answer: A) Improving data integrity and auditability

42. AI-driven security helps by —

A) Predicting and detecting threats
B) Decreasing encryption strength
C) Slowing down performance
D) Disabling firewalls
✅ Answer: A) Predicting and detecting threats

43. The most important factor for user trust in cloud services is —
A) Transparency
B) Branding
C) UI design
D) Cost
✅ Answer: A) Transparency

44. GDPR is a framework for —

A) Data Privacy and Compliance
B) Cloud Backup
C) API Management
D) Encryption Standards
✅ Answer: A) Data Privacy and Compliance

45. A key feature of Google Cloud’s security model is —

A) Default encryption for all data
B) No encryption
C) Optional encryption
D) Manual encryption only
✅ Answer: A) Default encryption for all data

46. Which of the following is NOT a privacy-preserving technique?

A) Anonymization
B) Differential Privacy
C) Homomorphic Encryption
D) Eavesdropping
✅ Answer: D) Eavesdropping

47. Virtualization security focuses on —

A) Isolating Virtual Machines (VMs)
B) File compression
C) Energy saving
D) None of these
✅ Answer: A) Isolating Virtual Machines (VMs)

48. A strong password policy includes —

A) Length + symbols + mixed case
B) Only numbers
C) Common words
D) User’s name
✅ Answer: A) Length + symbols + mixed case

49. Cloud monitoring tools help —

A) Detect anomalies and intrusions
B) Increase latency
C) Decrease performance
D) Disable encryption
✅ Answer: A) Detect anomalies and intrusions

50. The main theme of Week 10 module is —

A) Cloud Security and Privacy
B) Cloud Economics
C) Cloud Storage
D) Cloud Deployment Models
✅ Answer: A) Cloud Security and Privacy

Summary of Week-10: Clustering Algorithms

Page 1 of 3: Introduction & K-Means
 What is Clustering? It's an unsupervised learning technique for grouping similar
data points together. The goal is to maximize similarity within a group (intra-
cluster) and minimize similarity between different groups (inter-cluster).
 K-Means Clustering: A popular and simple algorithm.
 o Steps:
1. Randomly place K centroids (the centers of the clusters).
2. Assign each data point to its nearest centroid.
3. Move each centroid to the average (mean) location of all points
assigned to it.
4. Repeat steps 2 and 3 until the centroids stop moving.
o Pros: Fast and easy to understand.
o Cons: You must choose K beforehand, and it works best for round, evenly-
sized clusters. It's sensitive to the initial random placement of centroids.
 How to choose K? Use the Elbow Method: plot the sum of squared errors for
different values of K and pick the K where the graph forms an "elbow."
Page 2 of 3: Hierarchical, BIRCH, and CURE
 Hierarchical Clustering: Builds a tree of clusters (a dendrogram). You don't need to
choose K first.
o Approach: Starts with each point as its own cluster and repeatedly merges
the closest clusters until only one remains.
o How to measure "closest" clusters?
 Single Link: Distance between the closest points of two clusters. Can
create long, "chain-like" clusters.
 Complete Link: Distance between the farthest points of two clusters.
Tends to create compact clusters.
 BIRCH (For Large Datasets): Designed for huge datasets that don't fit in memory.
o Idea: It doesn't use the raw data. Instead, it summarizes the data into
"Clustering Feature" trees, which are then used for clustering. It's very fast
and memory-efficient.
 CURE (For Odd Shapes): Handles clusters that aren't round.
o Idea: Instead of just a centroid, it represents each cluster with multiple
representative points. These points are shrunk toward the center to reduce
the effect of outliers. This allows it to find clusters of arbitrary shapes.
Page 3 of 3: DBSCAN
 DBSCAN (Density-Based Clustering): Finds clusters based on dense regions of
points.
o Key Ideas:
 Core Point: A point with many neighbors (≥ minPts) within a small
radius (ε).
 Border Point: A point that has fewer neighbors but is close to a Core
Point.
 Noise Point: A point that is neither a Core nor a Border point (an
outlier).
o How it works: It starts with a point, finds all points reachable from it
(forming a cluster), and moves on to unvisited points.
o Pros: Doesn't require specifying the number of clusters (K). Can find clusters
of any shape and automatically identifies outliers (noise).
o Cons: Sensitive to the ε and minPts parameters. Struggles if clusters have
very different densities.

Summary of Week-11: Gaussian Mixture Models & Expectation-Maximization

Page 1 of 3: Introduction to GMMs
 Mixture Models: A way to model complex data by combining several simpler
probability distributions. Think of it as a "clusterer" that uses probabilities instead
 of hard boundaries.
 Gaussian Mixture Model (GMM): The most common type, where the simple
distributions are Gaussians (bell curves).
o Formula: p(x) = π₁*N(x|μ₁,Σ₁) + π₂*N(x|μ₂,Σ₂) + ...
o Components:
 π_k (Mixing Coefficient): The weight of each cluster (must sum to 1).
 μ_k (Mean): The center of the Gaussian for cluster k.
 Σ_k (Covariance): The shape (spread and orientation) of the Gaussian
for cluster k.
 Soft Clustering: Unlike K-Means, which forces a point into one cluster, GMM gives a
point a probability of belonging to each cluster. This is more flexible.
Page 2 of 3: The EM Algorithm
 The Problem: We don't know which cluster each point belongs to (this is the "latent
variable"), and we don't know the parameters (π, μ, Σ) of the Gaussians. It's a
chicken-and-egg problem.
 The Solution: Expectation-Maximization (EM) Algorithm. An iterative process that
solves this problem.
o Step 1 (E-Step - "Guess"): Using the current model parameters, calculate the
probability (responsibility γ) for each data point belonging to each cluster.
This is the "soft assignment."
o Step 2 (M-Step - "Maximize"): Using these new probabilities, update the
model parameters (π, μ, Σ) to best fit the data. This is like a weighted
version of calculating mean and variance.
 Repeat the E and M steps until the parameters stop changing significantly. The
algorithm is guaranteed to improve the model with each iteration, but it may only
find a good local solution, not the global best.
Page 3 of 3: Key Takeaways & Properties
 EM is a General Tool: While used here for GMMs, the EM algorithm is a powerful
framework for any problem with "missing data" or latent variables.
 GMM vs. K-Means:
o GMM allows for oval-shaped clusters (using covariance), while K-Means
assumes round clusters.
o GMM provides soft, probabilistic cluster assignments.
 Important Notes:
o Initialization Matters: EM can get stuck in a suboptimal solution, so good
starting points are important.
o It Estimates Parameters, not K: The EM algorithm for GMMs estimates the
weights (π), means (μ), and covariances (Σ). You must choose the number of
components (K) yourself.

Summary of Week-12: Learning Theory & Reinforcement Learning

Page 1 of 3: How Well Can a Model Learn? (Theory)
 Goal of Theory: To understand how hard a machine learning problem is and to give
guarantees about a model's performance on new, unseen data.
 Generalization Error: The true error of a model on the entire distribution of data.
This is what we care about but can't directly measure.
 Empirical Error: The error of a model on the training data. This is what we can
measure and try to minimize.
 Empirical Risk Minimization (ERM): The common practice of choosing the model
that has the lowest training error. The challenge is ensuring this model also has a
 low generalization error.
 Key Tool - Uniform Convergence: A theoretical guarantee that if the training set is
large enough, the training error for all models in our hypothesis class will be close
to their true error.
Page 2 of 3: VC Dimension and PAC Learning
 VC Dimension: A measure of the complexity or "flexibility" of a model.
o Shattering: A model can "shatter" a set of points if it can correctly classify
every possible way of labeling them.
o VC Dimension is the largest number of points that the model can shatter.
o Why it matters: A high VC dimension means the model is very complex and
can overfit (memorize the training data but fail on new data). A low VC
dimension means the model is simple and might underfit.
 Sample Complexity: The amount of data needed for a model to learn well. It grows
with the VC Dimension.
 PAC Learning (Probably Approximately Correct): A framework that defines what it
means for a problem to be "learnable."
o Probably (1-δ): The learning algorithm has a high probability of success.
o Approximately (ε): The model's error is small.
o A hypothesis class is PAC-learnable if an algorithm can, with high
probability, find a model that is approximately correct using a reasonable
amount of data. A finite VC dimension is a key requirement for PAC
learnability.
Page 3 of 3: Reinforcement Learning (RL)
 What is RL? A type of machine learning where an agent learns to make decisions by
taking actions in an environment to maximize a reward signal.
o It's NOT Supervised Learning (no labeled right/wrong answers).
o It's NOT Unsupervised Learning (no goal to find hidden structure).
 Key Concepts:
o Trial and Error: The agent tries different actions and learns from the
consequences (rewards/punishments).
o Delayed Reward: The feedback (reward) might come much later than the
action that caused it (e.g., winning a chess game).
o Exploration vs. Exploitation: The agent must balance trying new things
(exploration) to find better strategies, and using things it already knows
work well (exploitation).
 Example - Tic-Tac-Toe:
o State: The current board configuration.
o Action: Placing an X or O in an empty square.
o Reward: +1 for win, 0 for draw, -1 for loss.
o The agent learns a value function that estimates the probability of winning
from each state.