jmp242 wrote:
> 
> http://www.zenoss.com/product/case-studies/case-study-rackspace/ ?
> 


I can provide some general information.

Depending on the number of OIDs you are collecting for those 700 nodes, you may 
need to go with a distributed model. The important numbers to consider are the 
number of OIDs being collected and the frequency of your polling. A collector 
with 8 cores, 32GB of RAM, and 6 15K rpm drives in RAID10 will support around 
50,000 OIDs every 5 minutes and will have an average IOWait of around 20%. Much 
more load than that and we've found that IOWait spikes dramatically and the 
system becomes unstable/unresponsive. The Zenoss processes themselves don't 
really require much RAM, but having RAM available for buffering helps you coast 
through disk IO spikes. 

There are some storage tweaks you can apply to squeeze more performance out of 
the system.
 - Monitor the disk usage with an application like IOSTAT or SAR and adjust 
your RAID controller bias accordingly.
 - Mount your /opt/zenoss/perf directory on it's own RAID10 array, and set 
"noatime" in fstab for that mount
 - Mount /opt/zenoss/perf with 'data=writeback' if you are using ext3, you'll 
get a performance boost but you lose some reliabilty


Hope that helps some. If you can determine the number of OIDs and your polling 
frequency I may be able to give you a starting point for your hardware.


Chris




-------------------- m2f --------------------

Read this topic online here:
http://forums.zenoss.com/viewtopic.php?p=29095#29095

-------------------- m2f --------------------



_______________________________________________
zenoss-users mailing list
[email protected]
http://lists.zenoss.org/mailman/listinfo/zenoss-users

Reply via email to