So after some more experimentation yesterday, moving 15 GB files and such, the
lustre performance is fine.
It seems that the slow down occurs when a single file is being written in many
chunks, and not related to security at all. Could this be due to file locking?
Is there a good way to debug this?
Sincerely,
Don Kolva, PE, MCP
Director of Product Development
Accelerated Technology Laboratories, Inc.
Toll-free (US and Canada): 800-565-LIMS (5467)
Phone: 910-673-8165
Fax: 910-673-8166
www.atlab.com<http://www.atlab.com/>
"Innovation. Performance. Success."
[Description: cid:image002.jpg at 01C8D2EB.06A1E580]
From: lustre-discuss-bounces at lists.lustre.org [mailto:lustre-discuss-bounces
at lists.lustre.org] On Behalf Of Donald J. Kolva
Sent: Monday, September 12, 2011 10:22 AM
To: lustre-discuss at lists.lustre.org
Subject: [Lustre-discuss] Slow lustre after NIS
Lustre was as fast (or faster) than a local drive when security was handled by
local nodes on a cluster. Switching the cluster to NIS (and also switching the
MDS to the same NIS server) seems to have made Lustre 5x slower. Switching back
to "local" security does not restore the speed.
Can Lustre work with NIS? Does each OST need to be NIS or just the MDS?
Is there a good way to debug the performance of Lustre? I''ve poured
over the manual, but it doesn''t seem clear on how to debug this
performance problem.
Lustre over the 1 GB network is as slow as Infiniband.
Setup is on MDS, and three OST each running CentOS 5.5
Cluster is one headnode and 8 computenodes with RHEL 5.5
Thanks for any help!
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20110913/89e80dc1/attachment.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/jpeg
Size: 5417 bytes
Desc: image001.jpg
Url :
http://lists.lustre.org/pipermail/lustre-discuss/attachments/20110913/89e80dc1/attachment.jpe