Dear Kevin & John,
all the tuning is part of the startup script:
https://github.com/VI4IO/io-500-dev/blob/master/proposal-draft/dkrz/resul...
Basically, I tuned some parameters for ior-easy and setting the lustre
stripe counts.
The find finishes quickly because the now shipped find script is a
single node parallel version of find.
Also, the result does scale with the number of processes used and it
is (now) valid to use less directories of MD-easy to scan,
e.g. here 1000 directories are created but the first 100 are scanned.
Without such a handling, it is extremely difficult to run it on larger
systems (may take hours).
The IOOPs do not change, though even for smaller runs.
Regards,
Julian
2017-07-26 19:59 GMT+02:00 Harms, Kevin <harms(a)alcf.anl.gov>:
so is there a standard description of the system and tuning that is included with the
results?
kevin
________________________________________
From: IO-500 <io-500-bounces(a)vi4io.org> on behalf of Julian Kunkel
<juliankunkel(a)googlemail.com>
Sent: Wednesday, July 26, 2017 12:51:41 PM
To: io-500(a)vi4io.org
Subject: [IO-500] Results from DKRZ available
Dear all,
the first compliant results from DKRZ are now available on our
GIT-Dev:
https://github.com/VI4IO/io-500-dev/tree/master/proposal-draft/dkrz/resul...
We are looking for other sites to run the IO500 benchmark. If you are
interested best is to join the slack:
https://vi4io.slack.com/ to
discuss issues.
Regards,
Julian
--
http://wr.informatik.uni-hamburg.de/people/julian_kunkel
_______________________________________________
IO-500 mailing list
IO-500(a)vi4io.org
https://www.vi4io.org/cgi-bin/mailman/listinfo/io-500
--
http://wr.informatik.uni-hamburg.de/people/julian_kunkel