Optimizing Start-up Time of Serverless Applications

What is it about?

Serverless applications or Function-as-a-Service (FaaS) promise to ease application developers from having to manage serving infrastructure or scaling of their application, while offering a pay-by-use instead of paying for pre-allocated infrastructure. One of the major burdens of serverless is the so-called "coldstarts"; serverless platforms need to spin-up new VMs and initialize the execution environment after periods of inactivity or under high load. To tackle this issue, we present a tool that automatically optimizes the initialization phase of serverless applications. Our approach utilizes library debloating, which is a technique that removes redundant parts of libraries that otherwise would need to be executed during start-up of applications. Our system reduces monetary costs of serverless application by an average of ∼20% (cutting many applications’ costs by >50%) while improving end-to-end latency by up to 2 times and memory usage by up to 42%!

Featured Image

Photo by Growtika on Unsplash

Photo by Growtika on Unsplash

Why is it important?

Our tool achieves significant speedups (up to 2 times improvement of end-to-end latency) and monetary savings (average 20%, maximum > 50%) of serverless applications, while it can be used out of the box (no modifications of the application needed) for applications that target AWS Lambda serverless platform.