>> So far it looks good, but we still need to test all of AFP.
>
> I will do that today and report my findings.
The machine has been busy since yesterday running the distribution and
the AFP (except for slow) multiple times using different parameters.
The distribution is working fine.
But there
> But there are still (spurious) issues with some AFP sessions. With high
> contention (a total of 64 worker threads at the same time, e.g. with
> threads=2 and -j 32), some sessions will run into suspicious timeouts:
>
> Timing Median_Of_Medians_Selection (2 threads, 25.039s elapsed time,
> 44.46