Serverless Auto-scaling mechanism using Reinforcement learning

Sonawane, Vikrant Bhanudas (2023) Serverless Auto-scaling mechanism using Reinforcement learning. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration manual) Download (476kB) | Preview

Abstract

Serverless computing represents a paradigm of cloud computing that caters to the dynamic needs of users by providing computing resources on an as-needed basis, with charges being levied based on the actual usage of these resources. Infrastructure management is handled by cloud providers, thereby enabling developers to concentrate on business logic development. However, scalability management necessitates the optimization of resource provisioning based on workload, which can prove to be a daunting task and may impose administrative overhead. In order to overcome the challenges that arise when managing scalability for serverless applications, this paper delves into the use of Reinforcement Learning (RL) techniques for autoscaling mechanisms. To manage dynamic workloads while ensuring Quality of autonomous Service (QoS) guarantees and optimizing resource utilization, RL environments and agents are employed, utilizing Q-learning algorithms. Nonetheless, Q-learning algorithms are not without their constraints, such as the overestimation of action values and delays in training and action enforcement. To surmount these limitations, this research proposes the use of Double Q-learning as an alternative solution. The research is driven by a dual motivation. Firstly, it aims to assess the effectiveness of the Double Q-learning algorithm and investigate the feasibility of reducing enforcement time. Secondly, it aims to evaluate the performance of the developed agents in dealing with new workloads comprising multiple serverless applications and cloud services. The proposed mechanisms undergo evaluation in both real and simulated environments, leveraging the Knative open-source serverless platform. Their efficacy in efficiently managing scalability for serverless applications is subsequently validated.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Mijumbi, Rashid UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > QA Mathematics > Algebra > Algorithms > Computer algorithms
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Tamara Malone
Date Deposited:	21 Oct 2024 12:00
Last Modified:	21 Oct 2024 12:00
URI:	https://norma.ncirl.ie/id/eprint/7106

Actions (login required)

View Item