[lustre-discuss] IOR input for pathologic file system abuse
Glenn Lockwood
glock at lbl.gov
Fri Jun 2 16:48:42 PDT 2017
Hi Michael
Pathological is very subjective, but the standard IOPS test with IOR
involves doing 4K writes (or reads) with random but non-overlapping offsets:
srun -N 4 -n 128 ./ior -t 4096 -b 4096 -s 1024 -e -w -z
However IOR is not really designed to do truly terrible I/O patterns such
as those that involve re-writing the same blocks, so the best you can do
(without modifying IOR) is trigger a bunch of false sharing and lock
contention by issuing small, strided writes, e.g.,
srun -N 4 -n 128 ~/ior -t 1024 -b 1024 -s 4096 -e -w
You can salt to taste, for example, by using transfer/block sizes that
don't divide nicely into your stripe size. But what is it exactly that
you're trying to accomplish? Real applications, even the bad ones, usually
don't do the nicely defined patterns that IOR does.
Glenn
On Fri, Jun 2, 2017 at 12:32 AM, Michael Kluge <michael.kluge at tu-dresden.de>
wrote:
> Hi all,
>
> I am looking for IOR scripts that represent pathological use cases for
> file systems. Something like shared file access with a small, unaligned
> block size or random I/O to a shared file. Does anyone has some input for
> me there that he/she is willing to share?
>
>
> Regards, Michael
>
> --
> Dr.-Ing. Michael Kluge
>
> Technische Universität Dresden
> Center for Information Services and
> High Performance Computing (ZIH)
> D-01062 Dresden
> Germany
>
> Contact:
> Falkenbrunnen, Room 240
> Phone: (+49) 351 463-34217
> Fax: (+49) 351 463-37773
> e-mail: michael.kluge at tu-dresden.de
> WWW: http://www.tu-dresden.de/zih
>
>
>
> _______________________________________________
> lustre-discuss mailing list
> lustre-discuss at lists.lustre.org
> http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20170602/c8732da6/attachment.htm>
More information about the lustre-discuss
mailing list