Student Name
CS 7346
Date
Research and Report | Auto-scaling Containerized Cloud Applications
The paper “Auto-scaling containerized cloud applications: A workload-driven approach” details
several key contributions in cloud computing. Cloud computing has had a transformative
impact on business opportunities by offering technology such as elastic computing and
efficient pricing models. However, the authors noted a challenge in managing hardware
resources in cloud computing due to varying workloads and resource needs. To address the
challenge, the authors introduce the PACE framework: Performance-aware Auto-scaler for
Cloud Elasticity.
PACE takes a two-pronged approach: Reactive Auto-scaling, and Proactive Auto-scaling.
Reactive auto-scaling uses threshold-based rules to safeguard against application failures
during heavy workload periods, to help maintain system stability and availability during
unanticipated demand spikes. It monitors system metrics such as CPU usage, memory
demand, network input and output, and application-specific metrics. When the metrics breach
predefined thresholds, the reactive auto-scaler scales up to add resources or scales down to
reduce resources [1]. The advantage of reactive auto-scaling is its ability to promptly respond
to changes in the demand for resources. However, this approach relies on precise threshold
configurations. If the thresholds are incorrectly set, it can lead to premature or delayed scaling
which may cause performance bottlenecks and waste resources. Reactive auto-scaling helps
to maintain system stability by providing a safety net for applications during unexpected
heavy workloads.
Proactive Auto-scaling uses convolutional neural networks and K-means clustering to predict
future workload demands, and proactively adjust resources [1]. These techniques allow the
system to understand the complex workload and demand patterns of applications in order to
predict future demands accurately. Predicting heavier workloads ahead of time ensures that
the system is prepared for heavier demands when they occur. This helps manage resources
efficiently and increase or maintain user satisfaction by ensuring consistent performance and
avoiding disruptions to service. However, the proactive auto-scaling approach relies on
Student Name
CS 7346
Date
machine learning, time series analysis, and trend prediction, which requires a significant
amount of historical data to base predictions on. Furthermore, configuring and creating the
predictive models necessary for the approach requires a technical background and an
understanding of machine learning. By integrating both the reactive and proactive approaches,
PACE can provide a comprehensive solution using auto-scaling that responds to and
anticipates workload changes.
The authors also elaborate on the benefits of cloud computing, such as scalability and
elasticity. Scalability and elasticity allow systems to handle larger workloads and dynamically
adjust the available resources based on current demands. Cloud computing offers substantial
improvements over traditional computing solutions.
The paper also included an evaluation section that involved rigorous testing of the PACE
framework to evaluate its effectiveness, efficiency, and robustness. The authors subjected
PACE to a series of performance tests involving fake workloads to simulate real-world
applications.
Key performance indicators are response time, throughput, system latency, and error rates [1].
The comparative analysis between reactive and proactive auto-scaling is evaluated based on
each model. However, the balance between avoiding resource shortages and preventing
resource over- provisioning would be the focus of the study.
The authors also mention the cost-efficiency of resource management using cloud computing.
In order to maintain cost-efficiency, cloud users need to strike a balance between available
resources and their costs. PACE can help optimize this balance for cloud users. This helps
ensure a multifaceted approach for optimal application performance, minimizing costs,
reducing administrative overhead, and contributing to sustainable computing practices.
Auto-scaling solutions like PACE contribute to efficiency by dynamically adjusting resource
allocation in response to real-time application demands. This dynamism ensures that
applications receive sufficient resources to perform optimally during peak loads and conserve
resources during quieter periods. PACE's proactive auto-scaling, through predictive analysis,
prepares the system for anticipated load increases, potentially preventing the invocation of
higher-priced, on- demand instances. Its reactive component, meanwhile, ensures resources
Student Name
CS 7346
Date
are not wasted during low-usage periods. By scaling down resources when demand drops,
PACE ensures users don't pay for idle resources. Automated solutions like PACE reduce the
need for constant human monitoring and manual adjustments, lowering the overall
administrative overhead and the potential for human error. PACE is an efficient tool and an
effective solution to the challenges posed by auto-scaling with containerization in a cloud
computing environment.
Student Name
CS 7346
Date
References:
[1] Spyridon Chouliaras, et al. “Auto-Scaling Containerized Cloud Applications: A Workload-
Driven Approach.” Simulation Modelling Practice and Theory, Elsevier, 26 Sept. 2022,
[Link]/science/article/pii/S1569190X22001241?via%3Dihub.