No problem at the moment, just want to get your thoughts and/or recommendations on the performance of my cacti setup in my environment. When I add all of my devices into cacti, about what poll time should I expect? Any special tricks to reduce poll time (I have cactid and i have heard about boost)? Would playing around with process numbers and threads help (so far my tinkering has no effect)?
Environment:
medium-size university system WAN
50ish routers
1100+ switches (ranging from 16 to 48 ports)
200+ misc devices
Desired polling: I want everything to be 1min if possible, but I realize that more than likely this is not feasible. With this in mind, I am shooting for this:
routers (roughly 50): 1min
feed switches (roughly 150): 1min
everything else (1000+): 5min
cacti server:
Intel(R) Pentium(R) 4 CPU 3.40GHz
1GB RAM
FreeBSD 6.2 stable
cactid
cacti-0.8.6j
Currently, I only have 129 devices being graphed (roughly 10% of total devices). Below is the output of my polls:
06/13/2007 03:59:40 PM - SYSTEM STATS: Time:39.8632 Method:cactid Processes:1 Threads:15 Hosts:129 HostsPerProcess:129 DataSources:28933 RRDsProcessed:11535
This is acceptable at the current time, but when the rest of my hosts get added, I'm afraid the times will get greater than 1min.
Any input is greatly appreciated. Thanks in advance.
poller time question
Moderators: Developers, Moderators
- gandalf
- Developer
- Posts: 22383
- Joined: Thu Dec 02, 2004 2:46 am
- Location: Muenster, Germany
- Contact:
Re: poller time question
Number of ds/rrd's is required. Number of host's doesn't matter _that_ much. Most is SNMP, I assume, few scripts?matcraig wrote: Desired polling: I want everything to be 1min if possible, but I realize that more than likely this is not feasible. With this in mind, I am shooting for this:
routers (roughly 50): 1min
feed switches (roughly 150): 1min
everything else (1000+): 5min
cacti server:
Intel(R) Pentium(R) 4 CPU 3.40GHz
1GB RAM
FreeBSD 6.2 stable
cactid
cacti-0.8.6j
4 CPUs is good, are they DualCore?
More memory would be fine, but at least speed of disks is crucial.
Currently, there's a fix in upcoming rrdtool 1.3, that will reduce disk overhead. But quantities of improvement are not yet available.
Surely, the most challenging issue is the 1 min polling. 5 minutes should not be a problem. And Boost should help significantly
Reinhard
DataSources:28933 RRDsProcessed:11535 for 129 hosts so far. This is only around 10% of my network, and I expect the number of DS and rrds to be proportional to the number of hosts since i am collectingthe same data from mostly the same type of equipment. That would be around
300,000 DS and 115,000 rrds
Its all SNMP, no scripts
The cpu is only, a Pentium 4, not 4 cpus...
7200 rpm sata
300,000 DS and 115,000 rrds
Its all SNMP, no scripts
The cpu is only, a Pentium 4, not 4 cpus...

7200 rpm sata
Who is online
Users browsing this forum: No registered users and 3 guests