I was (am) the one running multiple performance monitors on the same server. Unfortunately, I haven't kept great notes, so I'm not sure if I did this before or after I discovered the "cachesize" option. That, and I had created multiple local zencommand instances because my rup(1) nagios-style plugin was timing out, and probably changed too many variables at once.
One of my servers is monitoring just over 4800 devices, but the configuration is very stripped down and is keeping only about 23,000 RRD files. I am running three instances of zenperfsnmp, but I'm beginning to test rolling all the devices back into a single zenperfsnmp instance with a cachesize of 50,000 (objects?). I'm also not running the following daemons: zenmodeler, zenperfxmlrpc, zenprocess, zensyslog, zenstatus, or zentrap. So your mileage will probably vary quite a bit. The hardware is a quad dual-core 3GHz Xeon with 10G of RAM and a single mirrored disk pair. On Tue, Jun 05, 2007 at 02:03:30AM +0000, cluther wrote: > chrisv, > Adding additional zenperfsnmp daemons on the same machine isn't > going to help your situation. zenperfsnmp is a very high performance > SNMP polling system, but it is almost always limited by the amount > of RAM on your system. Once your perf directory exceeds the size > of your available RAM (for file system caching) you're continually > invalidating cache and trashing the disk. > > So try to keep your > perf directory within the size of available RAM for disk caching. > This will keep zenperfsnmp humming along. -- David Carmean Network Appliance, Inc Infosystems Architect, 495 E. Java Drive Java (Sunnyvale) Engineering Lab Services Sunnyvale, CA 94089 _______________________________________________ zenoss-users mailing list [email protected] http://lists.zenoss.org/mailman/listinfo/zenoss-users
