Dynamic Prewarming Strategy using Reinforcement Learning and LSTM for Cold Start Mitigation in Serverless Computing

Kshirsagar, Sameer Nandkishor (2023) Dynamic Prewarming Strategy using Reinforcement Learning and LSTM for Cold Start Mitigation in Serverless Computing. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (3MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Function-as-a-Service (FaaS) is emerging as a transformative cloud computing paradigm, offering several advantages, including pay-per-use billing, rapid application deployment, and elastic resource management. Cold start delay is a major challenge in FaaS applications, where the initial invocation of a function can take significantly longer than subsequent calls due to the need to initialize the underlying function container. These delays can significantly impact user experience and overall system performance. To mitigate cold start delays, researchers have explored predictive prewarming techniques along with other solutions. Machine learning models like LSTM are used to predict upcoming function invocations patterns and prewarm containers accordingly. However, predictive prewarming can lead to excessive pre warming, consuming resources and increasing infrastructure costs. To overcome this limitation, this research presents a Hybrid dynamic prewarming strategy that leverages reinforcement learning (RL) and LSTM to optimize container prewarming decisions, striking a balance between cold start latency reduction and cost minimization. The RL agent continuously learns from historical function invocation patterns and current system conditions, making real-time prewarming decisions based on predicted demand and resource availability. Experimental evaluations demonstrate that the proposed strategy significantly improves traditional predictive prewarming approaches, achieving almost the same latency while minimizing resource costs by 30%. The dynamic nature of the RL-based approach ensures optimal prewarming decisions, effectively mitigating cold start delays while maintaining cost efficiency.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Makki, Ahmed UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QA Mathematics > Computer software T Technology > T Technology (General) > Information Technology > Computer software T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	08 Apr 2025 16:33
Last Modified:	08 Apr 2025 16:33
URI:	https://norma.ncirl.ie/id/eprint/7386

Actions (login required)

View Item