On 08/30/2013 03:40 PM, Al Viro wrote:
On Fri, Aug 30, 2013 at 03:20:48PM -0400, Waiman Long wrote:
There are more contention in the lglock than I remember for the run
in 3.10. This is an area that I need to look at. In fact, lglock is
becoming a problem for really large machine with a lot of cores. We
have a prototype 16-socket machine with 240 cores under development.
The cost of doing a lg_global_lock will be very high in that type of
machine given that it is already high in this 80-core machine. I
have been thinking about instead of per-cpu spinlocks, we could
change the locking to per-node level. While there will be more
contention for lg_local_lock, the cost of doing a lg_global_lock
will be much lower and contention within the local die should not be
too bad. That will require either a per-node variable infrastructure
or simulated with the existing per-cpu subsystem.
Speaking of lglock, there's a low-hanging fruit in that area: we have
no reason whatsoever to put anything but regular files with FMODE_WRITE
on the damn per-superblock list - the *only* thing it's used for is
mark_files_ro(), which will skip everything except those. And since
read opens normally outnumber the writes quite a bit... Could you
try the diff below and see if it changes the picture? files_lglock
situation ought to get better...
Sure. I will try that out, but it probably won't help too much in this
test case. The perf profile that I sent out in my previous mail is only
partial. The actual one for lg_global_lock was:
1.01% reaim [kernel.kallsyms] [k] lg_global_lock
|
--- lg_global_lock
mntput_no_expire
mntput
__fput
____fput
task_work_run
do_notify_resume
int_signal
|
|--51.62%-- __shmdt
|
--48.38%-- __shmctl
So it is the mnput_no_expire() function that is doing all the
lg_global_lock() calls.
Regards,
Longman
--
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
