Multimedia

Communication Systems

 From communication perspective, the higher layers of

Multimedia Communication System (MCS) is divided into two
architectural subsystems:
1. Application Subsystem
2. Transport Subsystem

On Application Subsystems, management and service issues

for group collaboration and session orchestration are
presented.
Group collaboration and session mngt provide support for a
large group of multimedia applns such as tele-collaboration.

 Transport subsystems includes a presentation of transport and

network layer protocols that are used for the standardized
support of networked multimedia applns.

Application Subsystem
Collaborative Computing the current infrastructure of
networked workstations and PCs and the availability of audio
and video at these end points, makes it easier for people to
cooperate and bridge space and time.
Network connectivity and end-point integration of multimedia
provides users with a collaborative computing environment.
3

 Collaborative computing is also known as Computer Supported

Cooperative Work (CSCW).
There are many tools for collaborative computing such as email, bulletin boards, screen sharing tools, text-based
conferencing systems etc.
Collaborative Dimensions electronic collaboration can be
categorized according to three main parameters time, user
scale and control.
Time - With respect to time, there are two modes of
cooperative work:
1. Asynchronous
2. Synchronous

 Asynchronous cooperative work specifies processing activities

that do not happen at the same time.
Synchronous cooperative work happens at the same time.
User scale the user scale parameter specifies whether a single
user collaborates with another user or a group or more than
two users collaborate together.
A group may be Static or Dynamic during its life time.
Group is static if its participating members are pre-determined
and membership does not change during the activity.
Group is dynamic if the no of group members varies during the
collaborative activity. Ie, group members can join or leave the
activity at any time.

 Group members may have different roles. Eg: member of a

group, participant of a group activity, conference initiator,
conference chairman, token holder or an observer.
Groups may consist of members which have homogeneous or
heterogeneous characteristics and requirements of their
collaborative environment.
Control can be centralized or distributed.
Centralized control means that there is a chairman who controls
the collaborative work and every group member reports to him
or her.
Distributive control means that every group member has
control over his/her own tasks in the collaborative work and
dbted control protocols are in place to provide consistent
collaboration.
6

Dimensions of Collaborative computing

 Other partition parameters may include locality and

collaboration awareness.
Locality partition means that a collaboration can occur either in
the same place or among users located in different places
through tele-collaboration.
Collaboration awareness divides group communication systems
into collaboration transparent, and collaboration aware
systems.
Collaboration transparent s/m is an existing application
extended for collaboration.
Collaborative aware system is a dedicated software application
for CSCW.

Group Communication Architecture

Group communication systems can be further categorized into
computer augmented collaboration systems, where
collaboration is emphasized, and collaboration augmented
computing systems, where the concentration is on computing.
Group communication (GC) involves the communication of
multiple users in a synchronous or an asynchronous mode with
centralized or distributed control.
Group communication architecture consists of a support
model, system model and interface model.
GC support model includes group commn agents that
communicate via a multi point multicast commn n/w.

10

 Group Rendezvous denotes a method which allows one to

organize meetings, and to get info abt the group, ongoing
meetings and other static and dynamic info.
Shared Applications denotes techniques which allow one to
replicate information to multiple users simultaneously.
Conferencing is a simple form of collaborative computing. This
service provides the mngt of multiple users for communicating
with each other using multiple media.
GC s/m model is based on a Client-Server model.
Clients provide user interfaces for smooth interaction btw group
members and the s/m.
11

 Servers supply functions for accomplishing the group commn

work, and each server specializes in its own function.
GC interface model includes two kinds of protocols for
exchanging info within the GC support model:
1. User presentation protocols
2. Group work management protocols
User presentation protocols perform interactions among the
clients such as opening a conference, closing a conference,
dynamic joining and leaving of a meeting etc.
Group work management protocols specify the communication
between the clients and the servers. Registration of active
conferences, queries for further conference info etc.
12

Group Rendezvous
These methods allow for setting up collaborative group
meetings and providing other static and dynamic info abt
groups and ongoing or future meetings.
Tools provide a single set of session and activity info at the user
interface.
Synchronous Rendezvous Methods these methods use
directory services and explicit invitations.
Directory services access info stored in a knowledge base abt
the conference.
Explicit invitations method sends invitations either point to
point or point to multipoint to conference participants.

13

 Asynchronous Rendezvous Methods these methods may be

implemented through email or bulletin boards.
Email based mechanism encapsulates in the body msg enough
info abt a group session establishment.
Bulletin boards are used to support asynchronous rendezvous.
These boards announce seminars, classes, conferences and
other open meetings of a school or institution.

14

Application Sharing Approach

Sharing applns is recognized as a vital mechanism for
supporting group commn activities.
When a shared appln pgm executes any i/p from a participant,
all execution results performed on the shared object are dbted
among all the participants.
Shared objects are displayed in shared window.
An important issue in appln sharing is shared control. To
determine whether they should be centralized or replicated.
Centralized Architecture - A single copy of shared application
runs at one site.
All participants i/p to the appln is forwarded to the local site
and the applns o/p is then dbted to all sites.
15

 Advantage easy maintenance bcoz there is only one copy of

the appln that updates the shared object.
Disadvantage high network traffic bcoz the output of the
appln needs to be distributed every time. A high bandwidth
network is needed.

16

 Replicated Architecture - A copy of the shared application runs

locally at each site. Input events to each appln is dbted to all
sites and each copy of the shared appln is executed locally at
each site.

Advantage low n/w traffic bcoz only i/p events are dbted among
the sites, low response times since all participants get their o/p
from local copies of the appln.
Disadvantage requires same execution envt for the appln at
each site, difficulty in maintaining consistency.
17

 An important problem with shared applns in a replicated

architecture is maintaining the consistency of shared objects.
The notion of a floor is used to maintain the consistency of
shared object data or applns shared among participants.
The group member who holds the floor has the right to
manipulate shared objects in shared window.
A CSCW control component resides at every site and dispatches
i/p events coming from an i/p device.
It checks if the active site is a floor holder. If so, it accepts and
processes the i/p event as well as distributes the i/p event to
other sites. If not, CSCW control discards its own i/p and accepts
i/p events coming from another site.
18

Conferencing
Conferencing supports collaborative computing and is also
called Synchronous tele-collaboration.
Is a management service that controls the commn among
multiple users via multiple media such as video and audio.
Conferencing services rely on low n/w latency for acceptable
user interactivity and high bandwidth for potentially data
intensive media.
They rely on dbted messaging for transmission of data and/or
control information.
Conferencing services control a conference. It include several
functions establishing a conference, closing a conference,
adding new users and removing users.
19

20

 Conference states can be stored either on a central machine

where a central appln acts as the repository for all information
released to the conference, or in dbted fashion.
Centralized Conference Control provides the establishment of
a conference.
First, the initiator starts a conference by selecting an initial
group of invited conference members. The knowledge of the
conference state is inquired from a central directory server.
Second, each invited client responds to the invitation so that
the initiator is informed to who will participate in the
conference.
A negotiation of conference policies and an admission of
resources is performed.
21

 During the negotiation, the shared conference state is

distributed using a reliable messaging service to all conference
participants.
All information related to the conference is stored on a central
machine.
This static control, implemented through explicit exchange of
the conference state, guarantees the consistency of the state
space to every participant and works well for small conferences.
Advantage guaranteed consistency of the conference state.
Disadvantage when a new participant wants to join, explicit
exchange of the conference state must be performed among all
participants, which causes large delays.
22

 Distributed Conference Control based on dbted conference

state.
Initiator of the conference establishes a multicast space with
multicast entries for dbn of info to the conference participants
and the conference is established.
Each site distributes its own participation status to other
conference participants, but there is no global notion of a group
membership and no guarantees that all users will have the
same view of the state space.
This loose control is implemented through retransmitting the
state periodically for eventual consistency.
It is done using an unreliable messaging service.
23

 The loose control is provided by sessions called lightweight

sessions, which are used on the Multicast Backbone (Mbone).
Mbone is a multicast capable segment of internet which has
been used for a number of applns including multimedia
conferencing.
Advantage Inherent fault tolerance, which means that if a n/w
connection breaks down, it is easier to re-establish the shared
conference.
Disadvantage conference participants may not have the same
view of the state space.

24

Session Management - Architecture

Session management is an important part of the multimedia
commn architecture.
It is the core part which separates control, needed during the
transport, from the actual transport.
Session mngt architecture is built around an entity session
manager, which separates the control from the transport.
Session control architecture consists of the following
components:
1. Session Manager
2. Media Agents
3. Shared Workspace Agent
25

26

Session Manager
Session manager includes local and remote functionalities.
Local functionalities may include,
Membership control mngt participant authentication etc
Control mngt floor control
Media control mngt intercommn among media agents or
synchronization
Configuration mngt exchange of interrelated QoS
parameters
Conference control mngt establishment, modification and
closing of conference.
The session mngr communicates with other session managers
to exchange session state info.
27

Media Agents
Responsible for decisions specific to each type of media.
Each agent performs its own control mechanism over the
particular medium.

Shared Workspace Agent

These agent transmits shared objects among the shared
application.

28

Session Management - Control

Each session is described through its session state.
State info (start time of session, policies associated with the
session, session name) is either private or shared among all
session participants.
Session mngt includes two steps to process the session state: an
establishment and a modification of the session.
During the establishment, the session mngr negotiates, agrees
and sets the logical state of its own session. Further, it
negotiates, agrees and sets billing policy and other policies with
other session managers.
Further, it permits publishing a session, allowing others to
locate and join a session.

29

 Control mechanisms embedded in session mngt are,

Floor Control

Is employed to provide access to the shared workspace.

Its often used to maintain data consistency.
Each s/m makes decisions, such as the level of simultaneity and
granularity at which to enforce access control.
Applns use a floor-passing mechanism, which means that at any
time, only one participant has the floor.
The floor is handed off to another participant when requested.
To obtain the floor, the participant must explicitly take action to
signal a floor change.

30

 Floor control for real time video is frequently used to control

bandwidth usage.
Floor control mechanisms are low level means used to
implement floor policies.
A floor policy describes how participants request the floor and
how the floor is assigned and released.
A session wide floor holder is assigned by a chairman.
Conference Control
Employed for conferencing applns.
31

Media Control
Includes a functionality, such as the synchronization of media
streams.

Configuration Control
Includes a control of media quality, QoS handling, resource
availability and other s/m components to provide a session
according to user requirements.
Membership Control
Include services like invitation to a session, registration into a
session, modification of
32
membership during the session etc.

 For dbtn of the shared session control info among the session
managers, reliable messaging services or unreliable messaging
services with periodic refreshment can be used.
Goal is to provide a distributed messaging service with different
degrees of reliable delivery.

33

Transport Subsystem
User and Application Requirements
Multimedia applns impose new requirements for,
Substantial data throughput Audio and Video data resemble a
stream like behavior and they demand high data throughput.
Fast data forwarding diff applns require data movement
ranging from normal, error free data transmission. The faster a
commn s/m can transfer a data packet, the fewer packets need
to be buffered.
It leads to spatial and temporal resource mngt in the end s/ms
and routers/switches.
34

 Service Guarantees to achieve service guarantees, resource

mngt must be used.
Without resource mngt, multimedia s/ms cannot provide
reliable QoS to their users bcoz transmission over unreserved
resources leads to dropped or delayed packets.
Multicasting is important in terms of sharing resources.
Processing and Protocol Constraints
A typical multimedia appln does not require processing of audio
and video to be performed by the appln itself.
Data are obtained from a source and are forwarded to a sink.

35

 Requirements are satisfied best if they take shortest possible

path through the s/m.
Low level data streaming corresponds to proposals for using
additional new buses for audio and video transfer within a
computer.
It enables switch based data transfer architecture.
Protocols involve a lot of data movement bcoz of layered
structure.
Segmentation and reassembly also occur.
Some protocols use retransmission error-recovery mechanism.
n/w works in an Asynchronous transfer mode.

36

Transport Layer
Transport protocols provide the following functions:
1. timing information
2. semi reliability
3. multicasting
4. NAK (Non-Acknowledgment) based error recovery
5. Rate control
Internet Transport Protocols TCP (Transmission Control Protocol)
TCP provides a reliable, serial commn path, or virtual circuit btw
processes exchanging a full duplex stream of bytes.
Each process reside in an Internet host that is identified by an IP
Address.
37

 Each process has a no of logical, full-duplex ports through which

it can set up and use a full duplex TCP connections.
Multimedia applns do not always require full-duplex
connections for the transport of continuous media.
Eg: a TV broadcast over LAN requires a simplex continuous
media connection.
During the data transmission, TCP must achieve reliable,
sequenced delivery of a stream of bytes.
TCP makes use of retransmission on timeouts and positive ack
upon receipt of information.
Sequencing of data is also done.

38

 For multimedia, positive ack causes substantial overhead as all

packets are sent with a fixed rate.
Negative ack would be a better strategy.
TCP is not suitable for real time video and audio transmission
bcoz its retransmission may cause violation of deadlines which
disrupt the continuity of continuous media stream.
TCP was designed suitable for non real time reliable applns.
UDP (User Datagram Protocol)
UDP supports multiplexing of data grams exchanged btw pairs
of Internet hosts.
Offers only multiplexing and checksumming.
39

 It provides real time transport property.

Can be used as a simple, unreliable connection for medium
transport.
Not suitable for continuous media streams bcoz it does not
provide the notion of connections.
RTP (Real time Transport Protocol)
Is an end-to-end protocol providing n/w transport functions
suitable for applns transmitting real time data such as audio,
video or simulation data over multicast or unicast n/w services.
RTP has a companion protocol RTCP (RTP Control Protocol) to
convey info abt the participants of a conference.
40

 Provide functions like determination of media encoding,

synchronization, framing, error detection, encryption, timing
and source identification.
RTCP is used for the monitoring of QoS and for conveying info
abt the participants in an ongoing session.
The first aspect is done by QoS Monitor, which receives RTCP
msgs. This monitor estimates the current QoS for monitoring,
fault diagnosis and long term statistics.
The second aspect is used for loosely controlled sessions.
It does not provide mechanisms to ensure timely delivery of
data or guaranteed delivery.
RTCP header carries sequence nos to allow end s/m to
reconstruct the senders packet sequence.
RTP make use of ST-II or UDP/IP for the delivery of data.
41

XTP (Xpress Transport Protocol)

XTP integrates transport and network protocol functionalities.
It defines six service types: connection, transaction,
unacknowledged data gram, acknowledged data gram,
isochronous stream and bulk data.
The end user is rep by a context becoming active within an XTP
implementations.
Two contexts are joined together to form an association.
The path btw two XTP sites is called a route.
There are two types of XTP packets: Information packets which
carry user data and Control packets which are used for protocol
management.
For Flow control, XTP uses sliding window or rate based flow
control.
42

 If window based control is selected, the window size is

negotiated during the connection setup.
To advance the flow control window, XTP uses a combined
mechanism btw a cumulative ack and a selective ack with a runlength encoding.
Data packet retransmissions are triggered by the arrival of
status reports showing missing data.
Timer controls the duration of the response to the request.
After the timer expires and a status report was not received, a
new status report is issued, and XTP enters a synchronizing
handshake, where all further data transmission are halted until
the correct status is received.
XTP error control is primary a set of building blocks, known as
mechanisms from which error control policies can be
constructed.
43

Features

XTP provides a connection oriented transport and network

transmission.
Different transport services like connection mode,
connectionless mode and transaction mode are provided.
Flexible error management allows the turning off of the
retransmission mechanism, which is useful for multimedia
applns.
XTP has rate based flow control.

44

Problems

XTP was designed to be implemented in VLSI to achieve high

performance.
XTP constantly enters the synchronization handshake, which is
very undesirable in high speed networks.
XTP has a large header, which creates overhead.
Source identification and discrimination are missing in XTP.
Inter networking with other protocols does not work with XTP.

45

Tenet Transport Protocols

This protocol that support multimedia transmission was
developed by the Tenet Group at the university of California at
Berkeley.
Transport protocols in this protocol stack are,
1. Real time Message Transport Protocol (RMTP)
2. Continuous Media Transport Protocol (CMTP)
They run above Real Time Internet Protocol (RTIP).
RMTP provides connection oriented, performance guaranteed,
unreliable delivery of messages.
Connection mngt and reliable delivery are absent from this
protocol.

46

 The main functions of this protocol are flow control and

fragmentation and reassembly of messages.
CMTP is designed to support the transport of periodic n/w
traffic with performance guarantees.
RMTP and CMTP provide data and continuous media
transmission but they obey Real-time Channel Administration
Protocol (RCAP), which provides resource reservation,
admission and QoS handling.
Heidelberg Transport System (HeiTS)
Is a transport system for multimedia communication.
47

 It was developed at the IBM European Networking Center,

Heidelberg.
HeiTS provides the raw transport of multimedia over networks.
It uses Heidelberg Continuous media Realm, which is a real time
envt for handling multimedia data within workstations.

A Multimedia Enhanced Transport Service (METS)

METS is the multimedia transport service developed at the
University of Lancaster.
It runs on top of ATM networks.

48

 Provides an ordered, but non-assured, connection oriented

commn service and features resource allocation based on users
QoS specification.
It allows the user to select upcalls for the notification of corrupt
and lost data at the receiver, and also allows the user to renegotiate QoS levels.
It incorporates buffer sharing, rate regulation, scheduling, and
basic flow monitoring modules to provide diff services such as
guaranteed services with deterministic QoS, statistical QoS
bounds and best effort services.

49

Network Layer
The requirements on the n/w layer for multimedia transmission
are a provision of high bandwidth, multicasting, resource
reservation and QoS guarantees, new routing protocols and new
higher capacity routers.
Internet Services & Protocols IP (Internet Protocol)
IP provides for the unreliable carriage of datagrams from source
host to destination host.
IP properties include,
Type of Service IP includes identification of the service quality
through the Type of Service (TOS) specification.

50

 TOS specifies precedence relation & services such as minimize

delay, maximize throughput, maximize reliability, minimize
monetary cost and normal service.
TOS can only be used if the n/w into which an IP packet is injected
has a class of service that matches the particular combination of
TOS markings selected.
Different classes may support diff media requirements.
Addressing & Multicasting Addressing means establishing a global
address space that allows every n/w in the internet to be
uniquely identified.
Addressing structure consists of 5 classes of address A, B, C, D, E
51

52

 LAN uses the concept of both broadcasting and multicasting.

Allows an appln to send a single msg to the n/w and have it
delivered to multiple recipients.
This concept is captured in the Internet Architecture through
special Internet class D addresses, in which multicast addressed
packets are routed to all targets that are part of the multicast
group.
Routing s/m must be made aware of which n/wks have hosts
participating in each multicast group, so that the arrival of a
multicast addressed packet can trigger proper forwarding to the
destination n/wks.
To avoid duplicative replications of multicast packets by multiple
routers, a spanning tree of routers is constructed.
53

 IP routing service has been provided through an Internet segment

called MBone (Multicast Backbone).
Mbone is a collection of UNIX workstations running a routing
daemon called mrouted, which is an implementation of the
Distance Vector Multicast Routing Protocol (DVMRP).
Mbone is layered on top of the Internets unicast topology.
Interconnectivity btw IP and underlying n/wks the mapping btw
IP and underlying n/w means the binding of IP addresses to lower
level n/w addresses.
Address Resolution Protocol (ARP) allows a router to broadcast a
query containing an IP address and receive back the associated
LAN address.
54

 Reverse Address Resolution Protocol (RARP) can be used to ask

which IP address is bound to a given LAN address.

Routing basic model of Internet consists of n/wks connected by

routers.
Asynchronous System (AS) are collection of routers falling under
a common administrative authority.
Routers commonly use the same routing protocol Interior
Gateway Protocol (IGP), within the AS.
AS of gateways exchange reachability info by means of an Exterior
Gateway Protocol (EGP).
Other protocols associated with routing are OSPF, BGP, IDRP etc.
55

Internet Group Management Protocol (IGMP)

IGMP is a protocol for managing Internet Multicasting groups.
It is used by conferencing applns to join and leave particular
multicast group.
Basic service permits a source to send datagrams to all
members of a multicast group.
There are no guarantees of the delivery to any or all targets in
the group.
Multicast routers periodically send queries (Host Membership
Query messages) to refresh their knowledge of memberships
present on a particular n/w.
56

 If no reports are received for a particular group after some no of

queries, the routers assume the group has no local members, and
they need not forward remotely originated multicasts for that
group onto the local n/w.
Otherwise, hosts respond to a query by generating reports (Host
Membership reports), reporting each host group to which they
belong on the n/w interface from which the query was received.
To avoid impulsion of concurrent reports, two possibilities are
used:
1. A host, rather than sending reports immediately, delays for
a D second interval the generation of the report.
2. A report is sent with an IP destination address equal to the
host group address being reported.
When a host joins a new group, it should immediately transmit a
report for that group, rather than waiting for a query, in case it is
the first member of the group.
57

Resource reSerVation Protocol (RSVP)

RSVP is a protocol which transfers reservations and keeps a
state at the intermediate nodes.
RSVP msgs are sent as IP datagrams, and the router keeps soft
state, which is refreshed by periodic reservation msgs. In the
absence of refresh msgs, the routers delete the reservation
after a certain timeout.
For implementing integrated services, four components are
needed packet scheduler, admission control routine,
classifier & reservation setup protocol.
RSVP protocol was designed to satisfy requirements like,
1. It must accommodate heterogeneous service needs.
2. It must give flexible control on reservation.
58

3. It must accommodate elementary actions such as adding

one sender/receiver or deleting one.
4. It must be robust enough to scale well to large multicasting
group.
5. It must provide for advance reservation of resources, and
for the preemption that this implies.
A Reservation specifies the amt of resources to be reserved for all
or some subset of the packets in a particular session.
Requires adding flow specification control state in the routers.
Resource qty is specified by the flow-spec, which parameterizes
the packet scheduling mechanism.
Packet subset to receive those resources is specified by filterspec, defines a packet filter that is instantiated in the classifier.
RSVP reservations are receiver oriented (sender starts but actual
reservation done by receiver)
59

60

Stream Protocol, Version 2 (ST-II)

ST-II provides a connection oriented, guaranteed service for
data transport based on the stream model.
The connection btw the sender and several receivers are setup
as uni-directional connections, although duplex connections can
be setup.
Is an extension of original ST Protocol. It consists of two
components:
ST Control Message Protocol (SCMP), which is a reliable,
connectionless transport for the protocol msg and ST Protocol
itself, which is an unreliable transport for the data.
ST-II provides a resource reservation during the connection
setup.
61

 The reservation is originated from the source.

It sends a SCMP msg with a flow specification, which describes
the stream requirements in terms of packet size, data rate etc.
If the destination accepts the call, it returns the final flow
specification to the source.
ST data packets do not carry complete addressing information.
They have a HOP IDentifier (HIP), similar to a virtual circuit
number.
ST-II is suitable for multimedia transmission bcoz of its resource
reservation along the path btw sender and receiver.

62

Real-Time Internet Protocol (RTIP)

RTIP provides for connection oriented, performance
guaranteed, unreliable delivery of packets.
It occupies an analogous place in the Tenet Protocol stack as the
IP in the Internet Protocol suite.
It communicates with RCAP for resource reservation, therefore
it provides guaranteed service.
Designed for real-time communication.

63

QoS and Resource Management

Basic Concepts
Parameterization allows for flexibility and customization of the
services.
It is defined in ISO standards through the notion of Quality of
Service (QoS).
QoS is defined as a concept for specifying how good the offered
networking services are.
It is a well-defined and controllable behavior of a system according
to quantitatively measurable parameters
QoS Layering
Traditional QoS was provided by the n/w layer of the
communication system.
64

65

66

 An enhancement of QoS was achieved through inducing QoS

into transport services.
MCS consists of 3 layers: Application, System and Devices.
System includes communication services and operating system
services. Devices include Network and Multimedia devices.

Service Objects
Services are performed on different objects media sources,
media sinks, connections and virtual circuits. Hence QoS
parameterization specifies these service objects.
In ISO stds, QoS description is meant to be for services,
processing a transport/network connection.
67

QoS Description
Application QoS parameters describe requirements for the
appln services possibly specified in terms of media quality,
which includes the media characteristics and their transmission
characteristics, such as end-to-end delay & media relations,
which specify the relations among media, such as media
conversion or inter/intra stream synchronization.
System QoS parameters describe requirements on the commn.
services and OS services resulting from the appln QoS. They
may be specified in terms of both quantitative and qualitative
criteria. Quantitative Criteria are those which can be evaluated
in terms of certain measures such as bits per second, number of
errors etc. Qualitative Criteria specify the expected services
needed for provision of QoS.
68

 Qualitative criteria can be used by the coordination control to

invoke proper services for particular appln.

Network QoS parameters describe requirements on network

services. They may be specified in terms of network load,
describing the ongoing n/w traffic and characterized through
average/minimal interarrival time on the n/w connection,
packet/cell size and service time in the node for the
connections packet/cell & network performance, describing
the requirements which the n/w services must guarantee.
Performance might be expressed through a source to
destination delay bound.
n/w services depend on traffic model and perform according to
traffic parameters.
69

QoS Parameter Values and Types of Services

The specification of QoS parameter values determines the types
of service like,
Guaranteed Services, Predictive Services & Best-effort
Services
Guaranteed Services provide QoS guarantees either in
deterministic or statistical repn.
Deterministic bounds can be given through a single value, a pair
of values or an interval of values.
It also deals with statistical bounds of QoS parameters.
Predictable Service based on past n/w behavior.
Best-effort Services based on either no guarantees or on
partial guarantees.
70

Resource Resource Management Architecture

A resource is a s/m entity required by tasks for manipulating
data.
Resources are active or passive. Can be either exclusively used
or shared. Can be single resource or multiple resource.

71

 The main goal of resource mngt is to provide guaranteed

delivery of multimedia data.
This goal implies 3 main actions :
1. To reserve and allocate resources during multimedia call
establishment.
2. To provide resources according to the QoS specification
[Link] adopt to resource changes during on-going multimedia
data processing.
The s/m includes Resource managers at the hosts as well as the
n/w nodes.
Resource management protocols are used to exchange infn abt
resources among the resource mngt.
72

73

Relation between QoS and Resources

QoS parameters specify the resource quantity allocated to the
services, as well as the service disciplines managing the shared
resource in MCS.
Relation btw QoS and resources is embedded in the form of diff
mappings btw QoS parameters and their corresponding
resources in resource mngt.
Client requests a resource allocation by specifying its
requirements. The Resource manager checks its own resource
utilization and decides if the reservation request can be served
or not.
74

Establishment and Closing of the Multimedia Call

Before any transmission with QoS guarantees in MCS can be
performed, several steps must be executed:
1. the appln (user) defines the required QoS.
2. QoS parameters must be dbted and negotiated.
3. QoS parameters btw diff layers must be translated if
their repn is different.
4. QoS parameters must be mapped to the resource
requirements.
5. required resources must be admitted/reserved/allocated
along the path btw sender(s) and receiver(s).
75

QoS Negotiation
The dbtn of QoS parameters requires two services - Negotiation
of QoS parameters and Translation of QoS parameters.
There are really two parties to any QoS negotiation: peer-topeer negotiation and layer-to-layer communication.
Peer-to-peer is also called caller-to-callee negotiation and layerto-layer negotiation is called service-user-to-service-provider
negotiation.
The purpose of negotiation is to establish common QoS
parameter values among the services users (peers) and service
providers (underlying layers).
76

77

 Bilateral Peer-to-Peer Negotiation this type of negotiation

takes place btw the two service users and the service provider is
not allowed to modify the value proposed by the service user.

78

 Bilateral Layer-to-Layer Negotiation this type of negotiation

takes place only btw the service user and the service provider.
This communication covers (1) btw local service users and
providers and (2) btw host sender and the network.
Unilateral Negotiation here, the service provider as well as
the called service user are not allowed to change the QoS
proposed by the calling user. This negotiation is reduced to
take it or leave it.
Hybrid Negotiation every participating host-receiver may
have diff capabilities from the host-sender, but still wants to
participate in the communication.
Bilateral layer-to-layer negotiation btw host sender and n/w,
Unilateral negotiation btw n/w and host receiver.
79

 Triangular Negotiation for Information Exchange in this

scheme, the calling user introduces into the request primitive
the avg value of a QoS parameter.
This value can be changed by the service provider/ callee along
the path through an indication/response primitive before
presenting the final value in the confirm primitive in the caller.

80

 Triangular Negotiation for a Bounded Target This is the same

type of negotiation as the previous one, only the values of a
QoS parameter are represented through two bounds:
target (average value)
lowest quality acceptable (minimal value)

The goal is to negotiate the target value, i.e., the service

provider is not allowed to change the value of the lowest
quality but is free to modify the target value.
The callee will make the final decision concerning the selected
value of the target.
This selected value of the QoS will be returned in confirm
primitive to the caller.
81

82

 Triangular Negotiation for a Contractual Value in this QoS

parameters are specified through a minimal requested value
and bound of strengthening.
The goal of this negotiation is to agree on a contractual value,
which in this case is the minimal request QoS parameter value.
The service provider can modify the minimal request value
towards the strengthening bound value.
The callee makes the final decision and reports with
response/confirm primitive to the caller.

83

84

 ST- II protocol provides end-to-end guaranteed service across

the Internet N/w.
The parameters related to the throughput are negotiated with a
triangular negotiation for a bounded target.
For parameters related to delay there is no negotiation.
The calling user specifies the maximum transit delay in the
connect request.
During the establishment of the connection, each ST-agent
participating in the stream will have to estimate the average
transit delay that it will provide for this stream and the average
variance of this delay.
The parameters related to the error control are not negotiated.
85

 RCAP (Real-time Channel Administration Protocol) and others

use triangular negotiation for diff QoS parameter values.
The QoS Broker is an end-to-end establishment protocol.
Includes bilateral negotiation at the application layer between
peers.
Unilateral negotiation with the OS.
Triangular negotiation at the transport subsystem layer with
underlying ATM network as the service provider.

86

87

Translation
It must always be possible to derive all QoS values from the user
and application QoS values.
This derivation -known as translation- may require 'additional
knowledge' stored together with the specific component.
Hence, translation is an additional service for layer-to-layer
communication during the call set-up phase.
Translation functions are as follows:
Human Interface Application QoS
Application QoS System QoS
System QoS Network QoS

88

Human Interface Application QoS

The service which may implement the translation between a
human user and application QoS parameters is called tuning
service.
A tuning service provides a user with an interface for input of
application QoS as well as output of the negotiated application
QoS.
The translation is represented through video and audio clips.
Application QoS System QoS
The translation has to map the application requirements into
the system QoS parameters, which may lead to translation such
as from 'high quality' synchronization user requirement to a
small (milliseconds) synchronization.
89

System QoS Network QoS

The translation maps the system QoS into the underlying

network QoS parameters and vice versa.
Important property of the translation service is that it must be
bidirectional translation.
Reverse translation results in Media Scaling. Media scaling
methods perform diff degrees of media quality degradation if
resources are not available.
Dynamic QoS change is used in conjunction with scaling
techniques.
90

Scaling
Scaling means to subsample a data stream and only present a
fraction of its original contents.
Can be done either at source or at the receiver. Diff methods are,
Transparent Scaling can be applied independently from the
upper protocol and appln layers. Usually achieved by dropping
some portions of the data stream.
Non-transparent Scaling implies a modification of the media
streams before it is presented to the transport layer. Requires
modification of some parameters in the coding algs or even
recoding of a stream that was previously encoded in diff format.
Scaling can be applied to both audio and video.
For audio, non-transparent scaling must be used.
91

For video different scaling can be applied.

Temporal scaling reduces the resolution of the video stream. The

no of video frames transmitted within a time interval decreases.
Spatial scaling reduces the no of pixels of each image in a video
stream.
Frequency scaling reduces the no of DCT coefficients applied to
the compression of an image.
Amplitude scaling reduces the color depth for each image pixel.
Color space scaling reduces the no of entries in the color space.

92

Resource Admission
After negotiation and translation, the next step is resource
admission.
It is the mechanism used to accept or reject a new connection.
It checks availability of shared resources using availability tests.
Resource availability tests are also called Admission tests.
Based on the result of these tests, the reservation protocol creates
either a reserve msg with admitted QoS values or a reject msg
when the minimal bound of QoS values cannot be satisfied.
Three types of tests are performed:
1. Schedulability test for shared resources
2. Spatial test for buffer allocation
3. Link bandwidth test for throughput guarantees
93

Resource Reservation / Allocation

Without resource reservation and management in end-systems
and routers/switches, transmission of multimedia data leads to
dropped or delayed packets.
The reservation of resources is in most systems is simplex, i.e., the
resources are reserved only in one direction on a link, which
implies that the senders are logically distinct from receivers.
The reservation of resources depends on the reservation model,
its protocols and a set of resource administration functions for
individual resources.
Resource reservation / allocation is done in 2 ways pessimistic
approach (make reservations for worst case) and optimistic
approach (make reservations according to an average workload).

94

Data structures & Functions for Resource Reservation

Resource Table contains information about the managed
resources. This includes static information and dynamic
information.
Reservation Table provides information about the connections
for which portions of the managed resources are currently
reserved.
Includes the QoS guarantees given to the connections and the
fractions of resource capacities reserved for these connections.

Reservation Function used during the call set-up phase,

calculates the QoS guarantees that can be given to the new
connection and reserves the corresponding resource capacities.
95

Reservation Model
There are three types of reservation model:
(1) Single Sender/Single Receiver (e.g., RCAP, QoS Broker);
(2) Single Sender/Multiple Receivers (e.g., ST-TI);
(3) Multiple Senders/Multiple Receivers (e.g., RSVP).
The reservation model is determined by its reservation direction
and style . The reservation direction can be sender-oriented (e.g.,
ST-11) or receiver-oriented (e.g., RSVP).
Sender-oriented reservation means that the sender transmits a
QoS specification (e.g., flow specification) to the targets. The
intermediate routers and targets may adjust the QoS specification
with respect to available resources before the QoS specification is
transmitted to the sender.
96

 Receiver-oriented reservation means that the receiver describes

its resource requirements in a QoS specification and sends it to
the sender in a 'reservation' message.
It is assumed that a sender has issued a 'path message before,
providing information about outgoing data.
The reservation style represents a creation of a path reservation
and time when the senders and receivers perform the QoS
negotiation and resource reservation.
The style for sender-oriented reservation may be either that the
sender creates a single reservation along the link to the receiver,
or the sender creates a multicast reservation to several targets.

97

 The style for receiver-oriented reservation is defined in RSVP as,

1. Wildcard- Filter style - a receiver creates a single
reservation, or resource 'pipe', along each link, shared among all
senders for the given session.
2. Fixed Filter style - each receiver selects the particular
sender whose data packets it wants to receive.
3. Dynamic Filter (DF) - each receiver creates N distinct
reservations to carry flows from up to different senders.

The reservation style can also be divided with respect to the time
when the actual resource allocation occurs: Immediate
Reservation & Advanced Reservation.
98

 The advanced reservation service is essential in multi-party

multimedia applications.
There are two possible approaches to the advanced reservation:
(1) a centralized approach where an advanced reservation
server exists,
(2) a distributed approach where each node on the channel's
path 'remembers' the reservations.

99

Resource Reservation / Allocation Protocols

A Resource reservation protocol performs no reservation or
allocation of required resources itself.
It is only a vehicle to transfer information about resource
requirements and to negotiate QoS values.
Resource reservation protocols are control protocols embedded in
multimedia call set-up (establishment) protocol.
The resource reservation protocol implies that every node and
host has a resource manager which is responsible for sending and
receiving the control messages, and invoking the resource
administration functions.
The resource manager works closely together with network
management agents for proper reservation and administration
decisions.
100

Working
The initiator of the connection (sender) sends QoS specifications
in a reservation msg (connect request).
At each router/switch along the path, the reservation protocol
passes a new resource reservation request to the resource
manager, which may consist of several components.
After the admission decision, the resource manager reserves the
resources and updates the particular service information for QoS
provision.

101

102

Resource Deallocation
After the transmission of media, resources must be deallocated
and buffer space must be freed and connections must be torn
down.
Step 1: The Sender requests closing of the multimedia call.
Resources for all connections corresponding to the multimedia call
along the path btw sender and receiver(s) must be deallocated
and the resource availability must be updated at every node.
Step 2: The Receiver requests closing of the multimedia call.
Request is sent to the sender and during the traversing of the
path, the resources are deallocated.

103

Managing Resources during Multimedia Transmission

There are several constraints which must be satisfied to provide
QoS guarantees:
Time constraints include delays
Space constraints include system buffers
Device constraints include frame grabbers allocation
Frequency constraints include network bandwidth for data
transmission
Reliability constraints

Rate Control
For MCS, rate based flow control and rate based service
disciplines are being introduced.
104

 The control mechanisms are connected with a connectionoriented n/w architecture which supports explicit resource
allocation and admission control policies.
Rate based service discipline is one that provides a client with a
minimum service rate independent of the traffic characteristics of
other clients.
It manages resources like bandwidth, service time and buffer
space.
Several rate based scheduling disciplines are developed like,
1. Fair Queuing
2. Virtual Clock
3. Delay Earliest Due Date (Delay EDD)
4. Jitter Earliest Due Date (Jitter EDD)
5. Hierarchical Round Robin (HRR)
6. Stop and Go
105

Fair Queuing
If N channels share an output trunk then each one should get
1/Nth bandwidth.
If any channel uses less bandwidth than its share, then this
portion is shared among the rest equally. This mechanism can be
achieved by the bit-by-bit round robin (BR) service among the
channels.
The BR discipline serves n queues in the round robin service,
sending one bit from each queue that has a packet in it.
Fair Queuing emulates BR as follows: each packet is given a finish
number, which is the round number at which the packet would
have received service, if the server had been doing BR.
The packets are served in order of the finish number.
106

Virtual Clock
This discipline emulates Time Division Multiplexing (TDM).
To each packet a virtual transmission time is allocated. It is the
time at which the packet would have been transmitted, if the
server would actually be doing TDM.
Delay EDD (Delay Earliest Due Date)
It is an extension of EDF scheduling (Earliest Deadline First) where
the server negotiates a service contract with each source.
The contract states that if a source obeys a peak and average
sending rate, then the server provides bounded delay.

107

 The server sets a packet's deadline to the time at which it should

be sent, if it had been received according to the contract. This
actually is the expected arrival time added to the delay bound at
the server.
By reserving bandwidth at the peak rate, Delay-EDD can assure
each channel a guaranteed delay bound.

Jitter EDD (Jitter Earliest Due Date)

Jitter EDD extends Delay-EDD to provide delay-jitter bounds.
After a packet has been served at each server, it is stamped with
the difference between its deadline and actual finishing time.
A regulator at the entrance of the next switch holds the packet for
this period before it is made
108
eligible to be scheduled.

Stop-and-Go
This discipline preserves the 'smoothness' property of the traffic
as it traverses through the network as follows:
The main idea is to treat all traffic as frames of length T bits,
meaning, the time is divided into frames.
At each frame time, only packets that have arrived at the server in
the previous frame time are sent.
Hierarchical Round Robin (HRR)
HRR server has several service levels where each level provides
round-robin service to a fixed number of slots.
Some number of slots at a selected level are allocated to a channel
and the server cycles through the slots at each level.
109

 The time a server takes to service all the slots at a level is called
the frame time at the level.
The key of HRR is that it gives each level a constant share of the
bandwidth.
'Higher' levels get more bandwidth than 'lower' levels, so the
frame time at a higher level is smaller than the frame time at a
lower level.
Weighted Fair Queueing (WFQ) Algorithm
Each packet is stamped with a time stamp as it arrives and then it
is transmitted in increasing order of the time stamps.

110

Traffic Shaping Schemes

Traffic shaping provides a set of rules that comply to the flow
traffic characteristics.
It should allow (1) a simple way for a connection to describe traffic
so that the n/w knows what kind of traffic to expect (2) a network
to perform admission control & (3) network to monitor
connection traffic and to confirm that the connection is behaving
as promised.
In Leaky Bucket, the sending host places the data into a bucket
and the data drain out at the bottom of the bucket being sent on
the n/w at a certain rate.
Rate is enforced by a regulator at the bottom of the bucket.
111

 Rate based disciplines are divided depending on the policy they

adopt.
Work-conserving Discipline serves packets at the higher rate as
long as it does not affect the performance guarantees of other
channels which also means a server is never idle when there is a
packet to be sent
Non-work-conserving Discipline does not serve packets at a
higher rate under any circumstances which also means that each
packet is assigned, explicitly or implicitly, an eligibility time.
Even when the server is idle, if no packets are eligible, none will be
transmitted.

112

End-to-End Error Control

Multimedia extensions to existing operating systems provide a fast
and efficient data transport between sources and destination
located at the same computer.
Degree of reliability is necessary bcoz of:
Decompression Technology Most audio and video compression
schemes cannot tolerate loss; they are unable to resynchronize
themselves after a packet loss or at least visible or / and other
perceptual errors are introduced
Human Perception Loss of digital audio is detected by a human
ear very quickly and results in lower acceptance of the multimedia
system.
Data Integrity in a recording appln, one cannot recover from an
error that is induced in the first recording of data.
113

 End-to-end error control consists of 2 steps: Error detection and

Error correction.
Error Detection

Reliability should be enforced, although there is some error

tolerance in multimedia systems.
This works, however, only if the application is able to isolate the
errors.
Existing error detection mechanisms such as checksumming and
PDU sequencing have to be extended towards conveying further
information.
These existing mechanisms allow detection of data corruption,
loss, duplication, and misorder at the lower levels.

114

 MPEG -2 encoded video is for enforcing error detection at a

higher layer.
This compression produces three type of frames in the video
streams.
I-frame contains the structural information of the video stream for
a certain time interval. P-frame and Bframe follow the I-frame
with supporting information.
In transport & lower layers, error detection mechanisms uses
lateness concept.
If a PDU arrives too late at the receiver, this info is useless for an
appln and should be detected as an error.
To identify late data, determine the lifetime of PDU and compare
their actual arrival time with their latest expected arrival time.
Protocol used for this kind of synchronization is Network Time
Protocol (NTP).
115

Error Correction
The traditional mechanism for reliability provision is
retransmission which uses either acknowledgment principle after
receiving data or window-based flow control.
If the acknowledgment is negative, the data is re-sent by the
sender.
The traditional reliable transfer strategies are not suitable for
multimedia communication bcoz of following reasons:
(1) with explicit ack, the amount of data to be stored at the sender
for potential retransmission can become very large.
(2) with the traditional window-based flow control, the sender
may be forced to suspend transmission while a continuous data
flow is required.
116

Error Correction
(3) the retransmitted data might be received 'too late' to be
consumed in real-time
(4) they are not designed for multicasting communication only for
point-to-point communication
Go-back-N Retransmission
if PDU i is lost, the sender will go back to i, and restart
transmission from i.
The successive PDUs after i are dropped at the receiver.
The lost PDU is recovered only if i <= n where n is specified at the
beginning of the transmission.
The receiver only send a negative acknowledgment if PDU i is lost.
117

Problems of Go-back-N
(1) Gap introduction & (2) Violation of throughput guarantees

Selective Retransmission
Provides better channel utilization.
The receiver sends a negative acknowledgment to the sender if
PDU i <=n is lost.
The sender retransmits only those PDUs which have been
reported missing, not the consecutive packets too.
Partially Reliable Streams
Introduce a weak concept of reliability.
Limits the no of packets to be retransmitted.
118

 Only the last n packets of the stream in a certain time interval will
be retransmitted.
Forward Error Correction (FEC)
Sender adds additional information to the original data such that
the receiver can locate and correct bits or bit sequences.
FEC mechanism can be specified by its code rate C (code
efficiency), which can be computed: C = S/ (S + E), S represents the
number of bits to be sent, E represents the number of added
check bits.
The redundancy introduced by the mechanism is (1 - C) and it
must be determined by the transport system.
Transport s/m needs two pieces of info: (1) error probability of the
n/w btw the sender and receiver & (2) the reliability required from
the appln.
119

 Low end-to-end delay and no need for exclusive buffering of data.

Also does not require a control channel from the receiver to the
sender.
Disadvantage FEC works only for error detection and correction
within a packet but not for complete packet loss.
Priority Channel Coding
Refers to a class of approaches that separates the medium into
multiple data streams with different priorities.
These priorities are then used to tag voice packets so that during
periods of congestion the network is more likely to discard low
priority packets which carry information less important for
reconstructing the original media stream.
Network switches drop a lower priority cells or provide a lower
grade of service during periods of network congestion.
120

Slack Automatic Repeat ReQuest (S-ARQ)

Error control scheme based on retransmission of lost voice packets
in high-speed LANs.
The packets are subject to delay-jitter, hence the receiver
observers gaps, which result in interruptions of continuous
playback of the voice stream.
Delay-jitter in packetized voice transmission is commonly
addressed through a control time at the receiver.
The first packet is artificially delayed at the receiver for the period
of the control time in order to buffer sufficient packets to provide
for continuous playback in the presence of jitter.
Slack time of a packet is defined as the difference between its
arrival time at the receiver and its playback time.
121

 Due to delay-jitter, a packet may arrive before or after its playback

time.
In the former case, the packet is placed in a packet voice receiver
queue until it is due for playback. In the later case, a gap has
occurred and the packet is played immediately.

The error control/correction schemes for multimedia

communication systems can be divided into two classes:
(1) partial retransmission mechanisms (e.g., Go-Back-N , Selective
retransmission, Partially Reliable streams, S-ARQ) and
(2) preventive mechanisms (e.g., FEC, Priority Channel Coding).

122

Resource Monitoring
This functionality is embedded in the resource manager and is
closely connected to the n/w mngt agent.
N/w mngt works with agents at every intermediate n/w node.
Agents are concerned with (1) info which can be exchanged
among the intermediate n/w nodes. Structure of this info is stored
in the Management Information Base. & (2) the protocol used to
exchange info btw the n/w agent and the managed component.
Monitoring in n/wks should be flexible, which means that
- most of the monitoring variables should be optional
- monitoring should be able to be turned on and off
There are 2 modes to operate resource monitoring end user
mode & network mode.
123

 End-user mode requests a status report abt the resources and

Network mode reports regularly the resource status on diff nodes
along the path btw communicating end users.
Monitoring at end-systems includes a supervisor function to
continuously observe if the processed QoS parameters do not
exceed their negotiated values.
Resource Administration Protocols provide communication about
resources between individual resource managers at the
intermediate nodes and end-points during multimedia
transmission.
They can be implemented either as part of the network
management protocols or as separate resource management
protocols.
124

Resource Adaptation
It is important for the MNS architecture to support dynamic
change of QoS parameters, so that they can be balanced to reach
an optimal value for all sessions in a predictable manner.
To achieve this goal, two factors must be provided.
(1) Notification & Renegotiation of QoS parameters
(2) Adaptive Resource Schemes
Renegotiation is a process of QoS negotiation when a call is
already set-up.
The renegotiation request can come either from the user, who
wants to change the quality of services, or from the host system
due to overload of the workstation or from the network due to
overload and congestion.
125

126

User Request for Renegotiation

If the user-sender requires a change of QoS, the renegotiation
entity has to check if local resources are available.
Further, the resource administration protocol has to be invoked to
check the availability of network resources if the change of QoS
requires change of network resources.
If resources are available, then resource reservation and allocation
is performed.
If the user-receiver requires a change of QoS, first the
renegotiation entity checks the local resource and reserves it.
Then the sender is notified via resource administration protocol,
and the same admission procedure follows.
At the end, the receiver has to be notified to change the local
resource allocation.
127

Host System Request for Renegotiation / Change

This request may come from the OS.
In this case, several users are admitted and some of the users
violate their admitted application requirements.
Then a notification about the degradation of the QoS performance
and renegotiation request occur.
The response is either adaptation of the misbehaved
user/application to the admitted level, or the misbehaved user's
acceptance of performance degradation.
This may also result in degradation of performance for other users
of the workstation which should be omitted by the OS control
mechanisms.

128

Network Request for Renegotiation / Change

Overload of the network at some nodes can cause a renegotiation
request for QoS change.
This request comes as a notification from the resource
administration protocol to the host reporting that the allocation of
resources has to change.

There are two possibilities:

(1) the n/w can adapt to the overload the n/w still needs to
notify the host bcoz some degradation may occur during the
modification of resources.
(2) the network cannot adapt to the overload the source
(host) must adapt.
129

Network Adaptation
The fixed routing and resource reservation for each conversation,
combined with load fluctuations, introduce problems of network
unavailability and loss of network management.
(1) increase network availability
(2) allow network administrators to reclaim resources
(3) reduce the impact of unscheduled, run-time
maintenance on clients with guaranteed services.
The more efficient the routing and the resource allocation is, the
greater the number of guaranteed connection requests is
accepted.
Different mechanisms like routing, performance monitoring,
dynamic re-routing and load balancing control must be
employed.
130

 Routing Mechanism implements the routing algorithm which

selects a route in adherence to certain routing constraints.
Performance Monitoring Mechanism monitors the appropriate
network performance and reports them to the load balancing
control.
Dynamic Re-routing Mechanism is needed to establish the
alternative route and to perform a transparent transition from the
primary route to the alternative route.
Load Balancing Control Mechanism receives information from the
performance monitoring mechanism and determines whether
load balancing can be attempted using a load balancing algorithm
defined by the policy.
131

Dynamic Re-routing Mechanism

When channel i is to be re-routed, the source tries to establish a
new channel that has the same traffic and performance
parameters and shares the same source and destination as
channel i, but takes a different route.
The new channel is called a shadow channel of a channel i.
After the shadow channel has been established, the source can
switch from channel i to the shadow channel and start sending
packets on it.
After waiting for the maximum end-to-end delay time of channel i,
the source initiates a tear-down message for channel i.
132

Source Adaptation
Adapt the source rate according to the currently available network
resources. Requires feedback information from the n/w to the
source which results in graceful degradation in the media quality
during periods of congestion.
Two possibilities to implement the feedback transmission
mechanism. (1) The per-connection state information is periodically
appended to a data packet for the corresponding connection. At the
destination, this information is extracted and sent back to the
source. A switch updates the information fields in a packet only if
the local service rate is lower than that reported by a previous
switch along the path. (2) The feedback message is sent in a
separate control packet which is sent back along the path of the
connection towards the source.
133

 Rate Control using N/w feedback each source adapts to changes

in n/w conditions caused by an increase or decrease in the no of
connections or by sudden changes in the sending rates of existing
connections.
Explicit and implicit feedback mechanisms are available.
Traffic shaping at Source obtained by smoothing over an interval
of 1- 4 frames.
Hierarchical Coding describes algs which produce two or more
types of cells describing the same block of pixels with different
degrees of detail.

134