[mpich-discuss] patch for ad_lustre_wrcoll.c

Martin Pokorny mpokorny at nrao.edu
Wed Aug 10 14:56:33 CDT 2011


I have modified ad_lustre_wrcoll.c to improve the performance of the 
ADIO Lustre code under workloads typical in my application. This patch 
should be correct in all cases, so I thought that I ought to submit my 
changes back to the code maintainers. The changes are relatively simple, 
  and I have not detected any errors in the files generated using this 
code; nevertheless, further testing in other applications would be in order.

These changes improve performance by reducing the number of system 
'write' calls in the ADIO Lustre collective write code, and perhaps also 
by keeping the writes ordered. This is especially effective in my 
application, in which the data are highly interleaved among the 
processes in the group calling the MPI-IO collective write functions.

It's difficult for me to get clean performance numbers with the system 
on which my application runs. However, by way of example, prior to these 
changes my application could not sustain 25 MB/s (on our 1 Gb 
Ethernet-connected, 4 OSS Lustre filesystem); with these changes 25 MB/s 
is no problem, and there are indications that 50 MB/s or better might be 
possible.

-- 
Martin

-------------- next part --------------
A non-text attachment was scrubbed...
Name: ad_lustre_wrcoll.c.diff
Type: text/x-patch
Size: 7072 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20110810/d6c90b8c/attachment.bin>


More information about the mpich-discuss mailing list