[Linux-cluster] Performance issue on our GFS cluster.

[Robert Moody - MWEB]

Hi there,
 
I am looking for any ideas as to why our cluster has a weird load issue.
We are currently running a 2 node GFS cluster with ES 4 and GFS 6.1.
 
Currently these systems are running a light load as a mail cluster. We
eventually plan to push our heavily loaded systems onto 3 node clusters.
However we are now concerend as the system load seems to be higher than
we expected it to be.
 
I have noticed that between the two nodes that load bounces between 0.5
and 1.8-2. It cycles between the machines. ie one will be 0.5 and then
the other would be 1.7 per say... then a few min later the one that was
not loaded is now loaded and the loaded one is now idle.
 
Our worrying concern is that the mail servers that we are going to move
onto these hosts are currently running a max of 0.5 however they process
a lot more of mail than the current GFS cluster.
 
The other day we did a patch update to see if any new patches corrected
these issues. However they are still there.
 
The top deamons in use are gfs_scand, gfs_inoded and dlm_recvd  However
they take about 7% of the CPU time which does not explain the high load
on the server. As I speak the once node has spiked to 2.98.
 
Any ideas?
 
Robert Moody
Snr. Sun Solaris/Linux Engineer
WPF : Communications Officer
MWeb
Tel : 021 596 8753
Cell : 084 466 8521
rmoody at mweb.com
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20060920/73510769/attachment.htm>