Performance observations and HOWTO

[Richard Purdie <richard.purdie@linuxfoundation.org>]

I was asked about whether we look at performance. We do have automated
monitoring but interpretation is left for humans. The basic reports are
linked from here:

https://autobuilder.yocto.io/pub/non-release/

under "Performance Reports" confusingly enough.

Taking a recent example:

https://autobuilder.yocto.io/pub/non-release/20230405-6/testresults/buildperf-alma8/perf-alma8_master_20230405030047_38e3769a72.html

We can see there has recently been:

a) parsing time increase from 10s to 22s
b) 3GB increase in TMPDIR size
c) 9MB rootfs size jump
d) 6 minute build time increase

The parsing time was probably my ptest change. For the build
time/tmpdir, it happened between revision 69808 and 69828.

You can resolve these with a poky repository:

git rev-list --reverse HEAD | nl -n ln | grep ^69808
git rev-list --reverse HEAD | nl -n ln | grep ^69828

and then look at the diff:

git log c7ee2d0236184045d46cfffa9f8822d3118baaf6..506e22a311cfe1136e70eadc0f1f07933aca072c

which shows it was likely the llvm/mesa-native changes.

Cheers,

Richard