I've been playing around with various Amazon EC2 instances lately. They offer a x1.32xlarge instance that features 128 vCPUs. So I did what any engineer with access to a corporate AWS account would do: I obtained an instance and measured how long it took to build Firefox!
$ time ./mach build ... Overall system resources - Wall time: 237s; CPU: 37%; Read bytes: 626688; Write bytes: 15316516864; Read time: 32; Write time: 2306432 real 3m59.590s user 172m42.840s sys 12m49.444s According to `dstat`, we do manage to saturate all available CPU cores and get 0% idle CPU for a large chunk of the "compile" tier. I think that's impressive. Even more impressive is there were no race conditions! The build system has come a long way. It's worth noting that a c4.8xlarge (which "only" has 36 vCPUs) can do a full build in 5 minutes, only a minute slower. Some of that is due to a higher clock speed. But most of that is due to the reality that there are still large chunks of the build where we can't saturate all available cores. In particular, there are long periods with low core utilization: * during configure * between the "export" and "compile" tiers (WebIDL and IPDL processing delay start of compile tier) * during libxul linking Just thought people would like to know how well C++ compiling scales in our build system these days. We've come a long way.
_______________________________________________ dev-builds mailing list dev-builds@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-builds
