[prev in list] [next in list] [prev in thread] [next in thread] 

List:       lustre-discuss
Subject:    Re: [lustre-discuss] Lustre-2.8.0 tunables
From:       "Dilger, Andreas" <andreas.dilger () intel ! com>
Date:       2016-06-03 6:47:00
Message-ID: B7C48A10-1530-4945-9E04-C557857D23BF () intel ! com
[Download RAW message or body]

On 2016/05/27, 13:29, "Ms. Megan Larko" <dobsonunit@gmail.com> wrote:
> 
> Greetings List!
> 
> I very recently installed Lustre-2.8.0 in CentOS 7.2 on bare metal
> in a small test environment for evaluation.
>
> I was going through my usual Lustre tunable process and I noticed
> that while there is still a /proc/fs/lustre/osc/{my_LustreOST-
> OSC_stuff}/max_rpcs_in_flight (=8), the corresponding item of
> max_dirty_mb (was =32 by default) is no longer present in
> Lustre-2.8.0.  There appears a new (to me) item of max_rpcs_in_progress
> (=4096 by default).  Would I still need to tune max_rpcs_in_flight
> now that I see this new max_rpcs_in_progress?   Is there a
> presentation/paper on Lustre-2.8.0 tunables (particularly for
> InfiniBand networks)?

AFAICS, there is still a max_dirty_mb on 2.8.0, both per-OSC and a
global limit:

$ rpm -q lustre-client
lustre-client-2.8.0-2.6.32_642.el6.x86_64.x86_64
$ lctl get_param max_dirty_mb
max_dirty_mb=7922
$ lctl get_param osc.*.max_dirty_mb
osc.myth-OST0000-osc-ffff88046630dc00.max_dirty_mb=32
osc.myth-OST0001-osc-ffff88046630dc00.max_dirty_mb=32
osc.myth-OST0002-osc-ffff88046630dc00.max_dirty_mb=32
osc.myth-OST0003-osc-ffff88046630dc00.max_dirty_mb=32
osc.myth-OST0004-osc-ffff88046630dc00.max_dirty_mb=32

But I don't see anything about "max_rpcs_in_progress"?

$ lctl list_param -R "*" | grep progress

Are you sure you have the stock 2.8.0 release, or is it a modified
version?

Cheers, Andreas
-- 
Andreas Dilger
Lustre Principal Architect
Intel High Performance Data Division


_______________________________________________
lustre-discuss mailing list
lustre-discuss@lists.lustre.org
http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org

[prev in list] [next in list] [prev in thread] [next in thread]