On Fri, Sep 2, 2022 at 16:21:51 +0200, Paul Gevers wrote: > > > On 11-07-2022 15:14, Julien Cristau wrote: > > > > What are the specs of these hosts? > > > > > > We have m5a.large instances with Amazon: > > > https://aws.amazon.com/ec2/instance-types/m5/ > > > > > > M5a and M5ad instances feature AMD EPYC 7000 series processors with an all > > > core turbo clock speed of 2.5 GHz. The AMD-based instances provide > > > additional options for customers that do not fully utilize the compute > > > resources and can benefit from a cost savings of 10%. With M5ad instances, > > > local NVMe-based SSDs are physically connected to the host server and > > > provide block-level storage that is coupled to the lifetime of the > > > instance. > > > > > > vCPU Memory (GiB) Instance Storage (GB) Network Bandwidth (Gbps) > > > EBS > > > Bandwidth (Mbps) > > > 2 8 EBS-Only Up to 10 Up to 2,880 > > > > > > > How long are tests allowed to take? > > > > > > 10000 seconds, i.e. 2 hours and 47 minutes. > > > > > Hmm. Looking at > > https://ci.debian.net/packages/m/mercurial/unstable/amd64/ some of the > > runs seem to finish in under 10 minutes, others take over 2 hours (but > > end up passing anyway). Are all instances the same? Most of the "fast" > > runs seem to be on ci-worker13. > > As I mentioned in my original report, I noticed the same about the passing > tests and I said ci-worker13 is a completely different beast. All instances > *except* ci-worker13 are the same, the one I mentioned above. ci-worker13 is > an m3.large.x86 host at equinix: > https://metal.equinix.com/product/servers/m3-large/ > How fast is the storage on the aws hosts? It looks like EBS has a number of variations with different characteristics, and the hg testsuite is probably quite dependent on that. It might also be interesting to see how it does when running in a tmpfs (or at least with /tmp on a tmpfs).
Cheers, Julien