Evaluating SigmaOS with Kubernetes for Orchestrating Microservice and Serverless Applications

Abstract

SigmaOS is a new multi-tenant cloud operating system that simplifies distributed application development. Its design centers around the novel concepts of realms and procs. A realm presents a tenant with a shared global namespace that hides the machine boundaries. Tenants structure their applications as process-like procs interacting through the realm’s namespace. Procs are lightweight, stateful, and can communicate. SigmaOS manages the scheduling and execution of procs to achieve high resource utilization and performance isolation.

This thesis compares SigmaOS with Kubernetes, a mainstream cloud operating system, using a microservice-style social network website and a serverless image resizing program. It measures their performances on a small-scale cluster in CloudLab. The SigmaOS version of the social network is easier to build (30% fewer lines), and its image resizing starts faster (25% - 89%). SigmaOS performs comparably to Kubernetes regarding latency and resource consumption when running a single application but provides better performance isolation when running multiple applications in separate realms: latency increases by 4-11% with concurrent applications in SigmaOS versus over 150% in Kubernetes.