Observium scales primarily on the number of interfaces you have, as they account for the majority of I/o. 128 switches would be a large install, 128 servers would not.
You should try running 2 or 3 pollers if your server's disk is not too heavily loaded. If you are only using one poller then your disk is waiting for SNMP was dating time.
Memcached was replaced with mysql memory tables, which you are already using :-)
Rrdcached seems to make little difference to me. I have my installs running in a ramdisk, as that removes the i/o bottleneck entirely.
Adam.
Tobias Brunner tobias.brunner@nine.ch wrote:
Hi,
I have some performance troubles with the poller, it takes a bit more than 5 minutes to finish. There are 128 devices to poll...
Looking at the "Performance tuning" wiki page and the mailing list, I implemented the following tweaks:
- Using poller-wrapper from Job Snijders ("Multiple poller instances" was
really bad, polling took about 10 minutes)
- Activated RRDcached
Putting the RRDs on a Ramdisk is not a solution for me (too less memory). Does memcache work with the current SVN build? What other options for performance tuning do I have? One idea was to have "worker" server which only poll a subset of devices and report the values back to the master. Is something like that possible? Which resources need the most I/O? RRD or MySQL? Are there many inserts and updates to the Database?
Thanks for all suggestions...
Cheers, Tobias
-- Nine Internet Solutions AG, Albisriederstr. 243a, CH-8047 Zuerich Support +41 44 637 40 40 | Tel +41 44 637 40 00 | Direct +41 44 637 40 13 Skype nine.ch_support _______________________________________________ observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium