> > Running SPECJbb2005. Higher bops are better. > > > > Kernel A = 4.18+ 13 sched patches part of v4.19-rc1. > > Kernel B = Kernel A + 6 patches > > (http://lore.kernel.org/lkml/[email protected]) > > Kernel C = Kernel B - (Avoid task migration for small numa improvement) i.e > > > > http://lore.kernel.org/lkml/[email protected] > > + 2 patches from Mel > > (Do not move imbalanced load purely) > > > > http://lore.kernel.org/lkml/[email protected] > > (Stop comparing tasks for NUMA placement) > > > > http://lore.kernel.org/lkml/[email protected] > > We absolutely need the 'best' pre-regression baseline kernel measurements as > well - was it > vanilla v4.17? >
I kept the baseline as 4.18. The only revert I know since 4.16 from a numa balancing front is the one where we unintentionally skipped task migrations. It did somehow give good results to a set of benchmarks but it would completely circumvent the task migration code. I am not sure if Jirka was taking the numbers from that kernel. From what I remember, we will pulled it out before 4.16 stabilized.
