Adrian Bunk <bunk@xxxxxxxxx> writes:
> Subject : x86_64: NR_IRQ increase causes 11.5% slowdown
> in lmbench's fork benchmark
> References : http://lkml.org/lkml/2006/11/2/192
> Submitter : Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx>
> Caused-By : Eric W. Biederman <ebiederm@xxxxxxxxxxxx>
> commit 550f2299ac8ffaba943cf211380d3a8d3fa75301
> Status : unknown
I'm not really certain who to cc on this one.
It appears the problem sequence is:
fork()
sched_setaffinity()
It got worse just because NR_IRQS doubled (and all of the associated
arrays doubled in size).
I haven't seen anyone reproduce this but Tim Chen, and Tim wasn't
able to root cause the problem so I believe we are going to have
this regression :(
Eric
|
