1) sounds like poor design to me<div><br></div><div>2) the sym links are removed and the files are copied from cache already</div><div><br><div class="gmail_quote"><div dir="ltr">On Tue, Jul 3, 2018, 08:23 John Price <<a href="mailto:jprice@csudh.edu">jprice@csudh.edu</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">If the data are being analyzed "as-is", then there will definitely be<br>
many small I/O calls. The batch system copies the files from tape to<br>
the /cache filesystem, and then makes a symlink to the disk on the local<br>
compute node. The way to avoid this is to copy the data files to the<br>
local disk before starting the analysis. <br>
<br>
I have a script that I have been using for this purpose with g12<br>
(granted, that's CLAS6, not CLAS12, but the principle is the same).<br>
<br>
John<br>
<br>
On Tue, 2018-07-03 at 08:03 +0900, Francois-Xavier Girod wrote:<br>
> The I/O to those jobs is defined as per CC guidelines, there is no<br>
> “small” I/O the hippo files are about 5 GB merging 10 evio files<br>
> together. I think the CC need to have better diagnostics. <br>
> <br>
> On Tue, Jul 3, 2018 at 7:56 AM Harout Avakian <<a href="mailto:avakian@jlab.org" target="_blank">avakian@jlab.org</a>><br>
> wrote:<br>
> <br>
> FYI<br>
> <br>
> I understood that was fixed. FX could you please check what<br>
> is the problem.<br>
> <br>
> <br>
> Harut<br>
> <br>
> -------- Forwarded Message -------- <br>
> Subject: <br>
> class12-2 jobs performing lots<br>
> of small i/o<br>
> Date: <br>
> Mon, 2 Jul 2018 08:58:04 -0400<br>
> (EDT)<br>
> From: <br>
> Kurt Strosahl<br>
> <<a href="mailto:strosahl@jlab.org" target="_blank">strosahl@jlab.org</a>><br>
> To: <br>
> Harut Avagyan<br>
> <<a href="mailto:avakian@jlab.org" target="_blank">avakian@jlab.org</a>><br>
> CC: <br>
> sciops <<a href="mailto:sciops@jlab.org" target="_blank">sciops@jlab.org</a>><br>
> <br>
> <br>
> Harut,<br>
> <br>
> There are a large number of clas12 jobs running through the farm under user clas12-2, these jobs are performing lots of small i/o.<br>
> <br>
> An example of one of these jobs is:<br>
> <br>
> Job Index: 55141495<br>
> User Name: clas12-2<br>
> Job Name: R4013_13<br>
> Project: clas12<br>
> Queue: prod64<br>
> Hostname: farm12021<br>
> CPU Req: 1 centos7 core requested<br>
> MemoryReq: 9 GB<br>
> Status: ACTIVE<br>
> <br>
> You can see the small i/o by looking: <a href="https://scicomp.jlab.org/scicomp/index.html#/lustre/users" rel="noreferrer" target="_blank">https://scicomp.jlab.org/scicomp/index.html#/lustre/users</a><br>
> <br>
> w/r,<br>
> Kurt J. Strosahl<br>
> System Administrator: Lustre, HPC<br>
> Scientific Computing Group, Thomas Jefferson National Accelerator Facility<br>
> _______________________________________________<br>
> Clas12_software mailing list<br>
> <a href="mailto:Clas12_software@jlab.org" target="_blank">Clas12_software@jlab.org</a><br>
> <a href="https://mailman.jlab.org/mailman/listinfo/clas12_software" rel="noreferrer" target="_blank">https://mailman.jlab.org/mailman/listinfo/clas12_software</a><br>
> _______________________________________________<br>
> Clas12_software mailing list<br>
> <a href="mailto:Clas12_software@jlab.org" target="_blank">Clas12_software@jlab.org</a><br>
> <a href="https://mailman.jlab.org/mailman/listinfo/clas12_software" rel="noreferrer" target="_blank">https://mailman.jlab.org/mailman/listinfo/clas12_software</a><br>
<br>
-- <br>
John W. Price<br>
Professor and Chair, CSUDH Department of Physics<br>
Coordinator, Science, Mathematics, and Technology Program<br>
Director, Office of Undergraduate Research, Scholarship, and Creative<br>
Activity<br>
310-243-3403<br>
<br>
</blockquote></div></div>