Bad poller performance on VM

[cacti_is_awesome]

Hi,
I've Cacti v0.8.7.e running on virtual machine with RHEL 5.7.

I've 4 vCPUs, 2.00GHz each and 16GB RAM.

RRAs are written to mounted NFS storage and monitoring about ~550 hosts.

Problem is the poller runtime which is been over 300 seconds and all my attempts to bring it down have failed so far.
I'm using spine.

Here's what I've already tried.

*Changed Process/Thread from 1/25 to 5/25, 3/25 and various other combinations, no luck with that.
*Checked NFS for slow response, errors, there aren't any.
*Changed poller process/thread counts, no luck.
*Added more CPUs, no luck there as well.

Can someone suggest how else I can tune up the performance? We'll be adding 400+ more servers soon to the same box.

Appreciate your response.

Thanks,
Ray

Here's snapshot of the settings/rrd I've.

09/28/2011 02:08:42 PM - SYSTEM STATS: Time:520.6432 Method:spine Processes:7 Threads:25 Hosts:545 HostsPerProcess:78 DataSources:15956 RRDsProcessed:11580
09/28/2011 02:09:12 PM - SYSTEM THOLD STATS: Time:0.0012 Tholds:0 Hosts:0
09/28/2011 02:17:33 PM - SYSTEM STATS: Time:452.5415 Method:spine Processes:7 Threads:25 Hosts:545 HostsPerProcess:78 DataSources:15956 RRDsProcessed:6861
09/28/2011 02:18:05 PM - SYSTEM THOLD STATS: Time:0.0012 Tholds:0 Hosts:0
09/28/2011 02:18:59 PM - SYSTEM STATS: Time:837.7530 Method:spine Processes:7 Threads:25 Hosts:545 HostsPerProcess:78 DataSources:15956 RRDsProcessed:12816
09/28/2011 02:20:08 PM - SYSTEM THOLD STATS: Time:0.0018 Tholds:0 Hosts:0
09/28/2011 02:24:51 PM - SYSTEM STATS: Time:589.1912 Method:spine Processes:7 Threads:25 Hosts:545 HostsPerProcess:78 DataSources:15956 RRDsProcessed:7682
09/28/2011 02:25:40 PM - SYSTEM THOLD STATS: Time:0.0012 Tholds:0 Hosts:0
09/28/2011 02:32:13 PM - SYSTEM STATS: Time:731.5647 Method:spine Processes:7 Threads:25 Hosts:545 HostsPerProcess:78 DataSources:15956 RRDsProcessed:10455

[bfek-18]

Hi,
The main problem here is the access via NFS.
Used the local disks.

bye
UK

[TheWitness]

You need to use boost first of all, and if using NFS, you need to use the rrdtool python wrapper to parallelize the RRDtool updates. It's on the forum. Search for it using google.

TheWitness