<br><tt><font size=2>mpich2-dev-bounces@mcs.anl.gov wrote on 08/21/2009
11:38:17 AM:<br>
> <br>
> When the workload is as you describe, ROMIO normally looks at the<br>
> accesses and if there is any overlap, it decides it would be better<br>
> served with independent access (in ad_write_coll.c there's a check
for<br>
> interleaved accesses). Contiguous data like your customer's
falls<br>
> under the non-interleaved category.<br>
> <br>
> However, on BlueGene, the romio_cb_read and romio_cb_write hints are<br>
> set to 'enable' instead of 'automatic'. This is usually the
right<br>
> thing, since aggregation works great on bluegene for workloads that<br>
> are non-overlapping, but also non-contiguous.<br>
</font></tt>
<br><tt><font size=2>Ok, so I tried hinting romio_cb_read/write to automatic
and it doesn't do the aggregation and performs much better. I like
that even better than hinting cb_buffer_size. </font></tt>
<br>
<br><tt><font size=2>I think we're ok. Like you said, we might want
to look at our defaults and selection logic eventually, but I think I'm
happy with my answer for now.</font></tt>
<br>
<br><tt><font size=2>Thanks</font></tt>
<br>
<br><tt><font size=2>Bob Cernohous: (T/L 553) 507-253-6093</font></tt>
<br>
<br><tt><font size=2>BobC@us.ibm.com</font></tt>
<br><tt><font size=2>IBM Rochester, Building 030-2(C335), Department 61L</font></tt>
<br><tt><font size=2>3605 Hwy 52 North, Rochester, MN 55901-7829</font></tt>
<br>
<br><tt><font size=2>> Chaos reigns within.</font></tt>
<br><tt><font size=2>> Reflect, repent, and reboot.</font></tt>
<br><tt><font size=2>> Order shall return.</font></tt>
<br>