On Mon, 2012-05-07 at 18:07 -0700, Eric W. Biederman wrote:
> "Paul E. McKenney" <paulmck@xxxxxxxxxxxxxxxxxx> writes:
> What I see as the real need is the filesystem modules need to do:
> rcu_barrier()
> kmem_cache_destroy(cache);
>
> Perhaps we can add some helpers to make it easy. But I think
> I would be happy today with simply moving the rcu_barrier into
> every filesystems module exit path, just before the file system
> module destoryed it's inode cache.
One liner kills the reap bottleneck and 99.999% of cache bloat. 1000
backgrounded vfstpd testcases finished ~instantly and left one
persistent pid namespace vs taking ages and bloating very badly.
Hacked up hackbench still hurts with all (except user) namespaces, but
that's a different problem (modulo hackbench wonderfulness).
Previous numbers:
default flags = SIGCHLD
-namespace: flag |= CLONE_NEWPID
-all: flags |= CLONE_NEWIPC | CLONE_NEWNET | CLONE_NEWUSER
marge:/usr/local/tmp/starvation # ./hackbench
Running with 10*40 (== 400) tasks.
Time: 2.636
marge:/usr/local/tmp/starvation # ./hackbench -namespace
Running with 10*40 (== 400) tasks.
Time: 11.624
marge:/usr/local/tmp/starvation # ./hackbench -namespace -all
Running with 10*40 (== 400) tasks.
Time: 51.474
New numbers:
marge:/usr/local/tmp/starvation # time ./hackbench
Running with 10*40 (== 400) tasks.
Time: 2.718
real 0m2.877s
user 0m0.060s
sys 0m10.057s
marge:/usr/local/tmp/starvation # time ./hackbench -namespace
Running with 10*40 (== 400) tasks.
Time: 2.689
real 0m2.878s
user 0m0.060s
sys 0m9.945s
marge:/usr/local/tmp/starvation # time ./hackbench -namespace -all
Running with 10*40 (== 400) tasks.
Time: 2.521
real 0m27.774s
user 0m0.048s
sys 0m21.681s
marge:/usr/local/tmp/starvation #
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
