> On 15. May 2026, at 15:12, Surya prakki <[email protected]> wrote:
> 
> Is the workload Read or Write intensive? If its read, whats the ARC cache hit 
> rate on the zfs side - if its not good, may be increasing ARC cache should 
> help; If its write, slog device helps as others already pointed out.
> Is the problem being seen at certain times (hour of the day/day of the week) 
> or such?
> Is the pool in general healthy? - no resilvering / failed drives !! 
> If all good on zfs side, is the client keeping enough OIO? b/w between the 2 
> is decent and getting utilized fully?
> I think a little more training is needed, I guess.
> my 2c.

1. grab zilstat and check if/how many sync writes you have (and plan for slog 
or set sync=disabled).
2. one mount does create one connection, if the workload has multiple 
threads/processes, use more mountpoints...

rgds,
toomas

> 
> On Wed, May 13, 2026 at 5:09 PM Lionel Cons <[email protected] 
> <mailto:[email protected]>> wrote:
>> On Wed, 13 May 2026 at 00:12, Dan McDonald <[email protected] 
>> <mailto:[email protected]>> wrote:
>> >
>> > On May 12, 2026, at 4:44 PM, Lionel Cons <[email protected] 
>> > <mailto:[email protected]>> wrote:
>> > >
>> > > On Tue, 12 May 2026 at 19:43, Dan McDonald <[email protected] 
>> > > <mailto:[email protected]>> wrote:
>> > >>
>> > >> On May 12, 2026, at 5:48 AM, Lionel Cons <[email protected] 
>> > >> <mailto:[email protected]>> wrote:
>> > >>>
>> > >>> Are there suggestions on how to DRASTICALLY reduce Illumos nfsd I/O
>> > >>> latency on ZFS?
>> > >>
>> > >> You're using a slog device on your pool, right?
>> > >
>> > > No, but I can try. Can slog devices be used for the root pool? Also,
>> > > can slog devices be added and removes dynamically?
>> >
>> > Yes, and yes, respectively.  HIGHLY recommend the lowest-latency device 
>> > you can get your hands on for a slog.  Think "small but fast NVMe drive."
>> 
>> Is there anything which can be done on the network and scheduler side?
>> 
>> Linux has realtime scheduler classes, could this be done for the nfsd
>> threads on Illumus too? Are there TCP6/IPv6 tuning parameters for
>> improved latency?
>> 
>> Lionel
>> 
>> ------------------------------------------
>> illumos: illumos-zfs
>> Permalink: 
>> https://illumos.topicbox.com/groups/zfs/T7c22b249653e43fb-Medd16ba86e5748e987f2e9be
>> Delivery options: https://illumos.topicbox.com/groups/zfs/subscription
> 
> illumos <https://illumos.topicbox.com/latest> / illumos-zfs / see discussions 
> <https://illumos.topicbox.com/groups/zfs> + participants 
> <https://illumos.topicbox.com/groups/zfs/members> + delivery options 
> <https://illumos.topicbox.com/groups/zfs/subscription>Permalink 
> <https://illumos.topicbox.com/groups/zfs/T7c22b249653e43fb-M129cf25301b34425a72d7e10>

------------------------------------------
illumos: illumos-discuss
Permalink: 
https://illumos.topicbox.com/groups/discuss/T7c22b249653e43fb-Md6eee6d7b5a8b484abb525a6
Delivery options: https://illumos.topicbox.com/groups/discuss/subscription

Reply via email to