Skip to content

Evaluate C3 and C3D machine types

Summary

In https://gitlab.com/gitlab-com/gl-infra/reliability/-/issues/18934, we changed to larger N2 instances.

We are however seeing forecasted CPU saturation in the future in https://gitlab-com.gitlab.io/gl-infra/tamland/patroni.html#patroni-ci-service-pg_primary_cpu-resource-saturation, and exploring ways to gain additional headroom. One way we can do this is by utilizing a larger instance type for our DB servers.

  • Current instance size: n2-highmem-128 - 128vCPU & 864GB RAM
  • Potential instance size: C3 @ 176vCPU & 1408GB RAM OR C3D @ 360vCPU & 2880GB RAM

We also previously evaluated N2D's as well. You can see the evaluation here: https://gitlab.com/gitlab-com/gl-infra/reliability/-/issues/17307

Proposal

Evaluate C3 and C3D machine types which offer:

  • About 50% more cores and RAM on Intel architecture with C3's
  • About 3x more cores and RAM on AMD EPYC 4th gen
  • Utilization of Google's new custom Intel Infrastructure Processing Unit which offloads work so more CPU can be used for workloads
  • Higher network and I/O capacity

Details: https://cloud.google.com/compute/docs/machine-resource

Edited by Nikolay Samokhvalov