pbojanic@clusterfs.com
2007-Jan-08 18:15 UTC
[Lustre-devel] [Bug 11302] liblustre scalability test
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11302
What |Removed |Added
----------------------------------------------------------------------------
CC| |mjmac@clusterfs.com
Status Whiteboard| |2007-01-08: scalability test
| |ready; CFS to determine plan
| |to exercise it and collect
| |results
Oleg, Jay, now that we have this liblustre scalability test, what are our next
steps to exercise it? Is this something that mjmac can evaluate on an XT3 at
Cray, or do we need a larger scale system like Rizzo at ORNL?
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=11302 (In reply to comment #5)> Oleg, Jay, now that we have this liblustre scalability test, what are our next > steps to exercise it? Is this something that mjmac can evaluate on an XT3 at > Cray, or do we need a larger scale system like Rizzo at ORNL?I''m wondering if the program isn''t expecting an evironment that doesn''t exist on an XT3: - it is compiled statically, so it can''t be run on Catamount nodes on an XT3. This leaves it running only on a linux node. - If run in two steps like shown, on a liblustre Catamount node, the state from the first test will be removed when the first run exits as Lustre mounts are per application invocation.
green@clusterfs.com
2007-Jan-12 07:37 UTC
[Lustre-devel] [Bug 11302] liblustre scalability test
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=11302 I am making jay''s comment public (perhaps it was made private by error). Also the whole purpose of this test program is to simulate large xt3 system on a single or two linux nodes. So one system emulates lots of connects and another (or same) is mds & ost(s). On xt3 no need for something like this, just create a program that opens/stats a file and waits, then run in on 10k nodes and you do not need anything else.