thold performance

[scotford]

I am running thold 3.2, Cacti 8.6i, Boost 1.2. Here are my stats without thold:
SYSTEM STATS: Time:37.0039 Method:cactid Processes:6 Threads:10 Hosts:572 HostsPerProcess:96 DataSources:55646 RRDsProcessed:0

When I add ~4000 thresholds my stats look like this:

SYSTEM STATS: Time:100.9239 Method:cactid Processes:6 Threads:10 Hosts:572 HostsPerProcess:96 DataSources:55646 RRDsProcessed:0

When boost commits the changes my stats look like this:
SYSTEM BOOST STATS: Time:252.7486 RRDUpdates:719244
SYSTEM STATS: Time:323.5080 Method:cactid Processes:6 Threads:10 Hosts:572 HostsPerProcess:96 DataSources:55646 RRDsProcessed:0

I am running MySQL 5.38 and I have converted all of my tables to InnoDB. The server has 4GB of RAM 2 Dual Core CPUs that are idle most of the time. One CPU core will peg at 100% with the mysqld as it is checking thresholds. The database and rrd files are on different LUNS on a DMX. The disk is fast enough right now that if I turn off boost my polling time increases by just a couple of seconds.

Is there anything else I can do to get thold to process the data faster or did I miss something?
Potentially I would like to get to ~20,000 thresholds. I am also going to add another 1000 hosts.

[harpermann]

Hi,
We're having similar issues. We are running a total of 46,500 thresholds. Cacti Thresholds can definitely handle that (the graphs are written perfectly, all inside of less than 3 minutes), but the thresholds take 5 hours to process, which makes them useless for monitoring purposes. The impact of this is not only that the thresholds take 5 hours to process, but also that the processing of the rrds for the graphs is also delayed by the threshold processing, so you dont get to see your graphs until the thresholds are done processing.

So how other people are doing with the Thold plugin?
What do you think the bandwidth is for the current Thold?
Any ideas for optimization? I will have resources to work on this since it's become an issue for us.

Regards,
- Harper

[cigamit]

Fixed a while back in SVN, there was a small bottleneck in the last version when you have a ton of thresholds.