Hi Teng,
that depends on over how many OSTs you stripe, and how many OSTs there
are per OSS. It is something you set up (default striping is all OSTs
with stripe size 1MB I believe) You need to get you stripe size and
stripe depth right (in the application, for example IOR) depending on
how much hardware you have.
There used to be some custom tools, but I think most use IOR nowadays (I
think) and do all kinds of test scenarios with that.
On 02/19/2015 03:26 PM, teng wang wrote:
Thanks for your advice Ron. The point is that I want to monitor how much
time each OSS spends on the data
transfer on the network and how much time it spends on accessing the disk
for a specific application. Is there any tool that can achieve such
dissection without
modifying the Lustre code?
On Thu, Feb 19, 2015 at 4:11 PM, Ron Croonenberg <[email protected]
<mailto:[email protected]>> wrote:
That depends on the number of IO nodes, compute nodes, lanes,
switches LNETS, OSSs etc etc. it is hard telling
On 02/19/2015 03:00 PM, teng wang wrote:
Thanks for all your answers. I just took some time to understand
how Obdfilter-survey works. It works fine for the Lustre local
filesystem.
But is there any Lustre tool that can directly profile the time
spent
on the network and on the disk for an application running at the
user level?
T
On Thu, Feb 19, 2015 at 10:47 AM, teng wang <[email protected]
<mailto:[email protected]>
<mailto:[email protected] <mailto:[email protected]>>> wrote:
Is there any way to benchmark the local filesystem
performance of
Lustre on the OSS side? For example, I want to benchmark
the random
and sequential I/O bandwidth of the local filesystem of
Lustre 2.5
using IOR.
Is there anyway to run IOR directly on its local filesystem
without
having
to go through the network?
Thanks!
Teng
_________________________________________________
Lustre-discuss mailing list
[email protected].__org
<mailto:[email protected]>
http://lists.lustre.org/__mailman/listinfo/lustre-__discuss
<http://lists.lustre.org/mailman/listinfo/lustre-discuss>
_________________________________________________
Lustre-discuss mailing list
[email protected].__org
<mailto:[email protected]>
http://lists.lustre.org/__mailman/listinfo/lustre-__discuss
<http://lists.lustre.org/mailman/listinfo/lustre-discuss>
_______________________________________________
Lustre-discuss mailing list
[email protected]
http://lists.lustre.org/mailman/listinfo/lustre-discuss
