[Bugs] [Bug 1467614] Gluster read/write performance improvements on NVMe backend

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1467614



--- Comment #56 from Manoj Pillai <mpillai at redhat.com> ---
(In reply to Mohit Agrawal from comment #55)
> Hi,
> 
>   I think we can try one way to limit cpu quota usage for fuse thread only.I
> am assuming if we do set cpu quota limit for fuse start thread then IOPS can
> improve because it could be happening other threads are not able to consume
> CPU(as we can see in previous output io thread
> is not consuming more CPU and fuse thread is consuming high CPU) because
> most of the time CPU is consumed by fuse start threads.
> 
>   If we are interested to try this quick and dirty way then I can provide
> options to use this.
> 
> Regards
> Mohit Agrawal

I'll paste the top -H output so that it shows overall cpu utilization as well.
us+sys is <32%; there is plenty of CPU cycles for other threads. The problem is
that the fuse thread has too much "work" -- either real work or spinlock
contention -- so it becomes the bottleneck. limiting its cpu usage doesn't seem
like the right thing to do.

<quote>
top - 00:09:30 up 13 days, 17:47,  4 users,  load average: 37.42, 15.02, 5.71
Threads: 371 total,  12 running, 359 sleeping,   0 stopped,   0 zombie
%Cpu(s): 20.0 us, 11.7 sy,  0.0 ni, 28.4 id, 38.9 wa,  0.0 hi,  1.1 si,  0.0 st
KiB Mem : 49279472 total, 48222328 free,   604892 used,   452252 buff/cache
KiB Swap: 24772604 total, 24772604 free,        0 used. 48025148 avail Mem

  PID USER      PR  NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND
24932 root      20   0 1284136  27576   3680 R 97.9  0.1   4:08.73 glusterfuse+
24925 root      20   0 1284136  27576   3680 R 72.5  0.1   2:26.06 glusterepol+
24928 root      20   0 1284136  27576   3680 S 72.5  0.1   2:26.01 glusterepol+
24927 root      20   0 1284136  27576   3680 R 72.4  0.1   2:25.79 glusterepol+

</quote>

-- 
You are receiving this mail because:
You are on the CC list for the bug.
Unsubscribe from this bug https://bugzilla.redhat.com/token.cgi?t=eMVlfS8aJs&a=cc_unsubscribe