SOCK: Serverless-Optimized Containers

Serverless computing is becoming increasingly popular as a way to avoid paying for idle periods and gracefully handle load spikes. Serverless platforms typically use containers to isolate lambda instances. General-purpose container systems such as Docker, however, are not well suited to serverless sandboxing and introduce unnecessary startup costs. In this work, we analyze the tradeoffs offered by alternative containerization primitives and use our findings to build a lean container system, SOCK, optimized for serverless workloads. Replacing Docker with SOCK in the Open-Lambda serverless platform results in an 18x speedup.