TechPrufer
Get started
Cloudflare
Cloudflare logoCloudflare · Feature trail

Go Garbage Collection Tuning

2 posts · Aug 2015 – Nov 2017

What this trail covers

Cloudflare's Go garbage collection tuning has evolved to optimize performance on multi-core machines by adjusting the GOGC variable. This post details how tuning GOGC can significantly improve throughput in scenarios with high allocation rates, such as benchmarks generating large amounts of short-lived data, leading to better performance scaling on many-core CPUs.

All Cloudflare trails
  1. Go, don't collect my garbageThis post details an investigation into the performance impact of Go's garbage collector (GC) on multi-core machines. It demonstrates how the default GOGC setting leads to frequent and expensive GC cycles when generating large amounts of short-lived data, significantly degrading performance. The post shows that by tuning the GOGC variable, performance can be dramatically improved, achieving a 22.56X speedup compared to the single-core score in the benchmarked ECDSA P256 signing operation. It also highlights the importance of understanding GC behavior and tuning it for specific workloads.
  2. Quick and dirty annotations for Go stack tracesThis post introduces a technique for improving the debuggability of Go stack traces by converting IPv4 addresses to their uint32 representation and passing them as blank identifier parameters to goroutine functions. This allows for easier identification of listening IPs within stack traces, especially in high-concurrency environments like Cloudflare's RRDNS server.