Performance Evaluation of Snapshot Methods to Warm the Serverless Cold Start

TL;DR

This paper evaluates snapshot-based methods, Prebaking and SEUSS, for reducing cold start delays in serverless computing, demonstrating significant performance improvements with different function complexities.

Contribution

It provides a comparative analysis of two snapshot techniques, Prebaking and SEUSS, highlighting their trade-offs and effectiveness in mitigating cold start latency.

Findings

Prebaking outperforms SEUSS in startup time for simple functions.

Prebaking reduces Markdown function first request time by 69%.

Performance gains vary with function complexity.

Abstract

The serverless computing model strengthens the cloud computing tendency to abstract resource management. Serverless platforms are responsible for deploying and scaling the developer's applications. Serverless also incorporated the pay-as-you-go billing model, which only considers the time spent processing client requests. Such a decision created a natural incentive for improving the platform's efficient resource usage. This search for efficiency can lead to the cold start problem, which represents a delay to execute serverless applications. Among the solutions proposed to deal with the cold start, those based on the snapshot method stand out. Despite the rich exploration of the technique, there is a lack of research that evaluates the solution's trade-offs. In this direction, this work compares two solutions to mitigate the cold start: Prebaking and SEUSS. We analyzed the solution's performance with functions of different levels of complexity: NoOp, a function that renders Markdown to HTML, and a function that loads 41 MB of dependencies. Preliminary results indicated that Prebaking showed a 33% and 25% superior performance to startup the NoOp and Markdown functions, respectively. Further analysis also revealed that Prebaking's warmup mechanism reduced the Markdown first request processing time by 69%.