A Deep Learning Based Framework to Initialize New Containers and Reduce Cold Start Latency in Serverless Platforms

Govindan, Surya Kumar (2020) A Deep Learning Based Framework to Initialize New Containers and Reduce Cold Start Latency in Serverless Platforms. Masters thesis, Dublin, National College of Ireland.

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

Abstract

Serverless computing has recently attracted large significance in the field of Cloud Computing because of its offerings like zero administration, infinite & automatic scaling and fine-grained billing to customers. However, this comes at the expense of having to deal with an unavoidable performance issue - the cold-start latency while initializing new containers. This is an important problem that cannot be ignored as it causes lags and delays in function execution, leaving the platform unsuitable for latency-sensitive applications. To efficiently avoid cold-start latency, this paper proposes a Deep Learning based Serverless (DLS) framework which initializes and sets up containers before the function execution request arrival, as opposed to initializing containers when or after the function execution request arrival. The DLS also makes use of the cache to store and reuse runtime libraries to minimize or avoid the time taken to download and install libraries. The framework provides at least 1.8x times faster function execution times, reduces memory usage up to 33%, and decreases CPU utilization up to 9.5% for applications with recognizable usage patterns, when compared to platforms like Apache OpenWhisk. As platforms like AWS Lambda have started investing towards predictable prewarming of containers, it becomes significant to explore using machine learning in serverless platforms. The reduction in cold start latency benefits both customers and industries in terms of billing and improving resource efficiency respectively. However, the proposed framework’s efficiency on user-interaction or dialog-based applications is yet to be analysed.

Item Type:	Thesis (Masters)
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
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Dan English
Date Deposited:	28 Jan 2021 14:03
Last Modified:	28 Jan 2021 14:03
URI:	https://norma.ncirl.ie/id/eprint/4535

Actions (login required)

View Item