Efficient Asynchronous RPC Calls for Microservices: DeathStarBench Study

TL;DR

This paper identifies thread management bottlenecks in asynchronous RPC calls within microservice benchmarks and demonstrates that replacing threads with fibers can significantly improve throughput and latency performance.

Contribution

It introduces a fiber-based implementation for asynchronous RPC calls in DeathStarBench, significantly enhancing throughput and reducing latency.

Findings

Peak throughput increased by up to 6x

Latency remained stable at high request rates

Thread management bottleneck was effectively mitigated

Abstract

Crucial in the performance of microservice applications is the efficient handling of RPC calls. We found that the asynchronous call implementation in a popular microservice benchmark suite, DeathStarBench, causes a bottleneck in thread management, reducing the peak throughput and increasing latency at high request rates. Replacing the threaded implementation with a fiber implementation increases peak throughput by up to 6x, meaning that the service can operate at a higher request rate without significantly increasing response latency.