Cacti performance OBSD vs. CentOS

[ai-danno]

I'm trying to get all the Linux out of my life, but the one sticking point still getting to me is my Cacti installation. Time and again, CentOS works fine in terms of response and performance for Cacti while OBSD runs it horribly slow.

I've tried tweaking MYSQL and Apache, but to no avail.

I was just wondering if anyone else who runs Cacti has been able to do so successfully on OBSD.

And... no, it's not a hardware issue. I've literally used the same piece of hardware with both OS's and gotten the same devastating results.

[scottro]

I've only run it on CentOS. I haven't even tried on FreeBSD, let alone Open.
Not that it's much help, just to let you know there's another Cacti person here.

[ai-danno]

Actually I appreciate you mentioning, I was wondering if I was the only one who used it.

[scottro]

It's an interesting program. Basically, my boss said learn it and gave me all the time I needed. I'm still stuck on some things--some of the docs amuse me. For instance, there's one that talks about some configuration or another, then says see the whatever configuration part of this document. That would be fine save that that part of the document doesn't exist.

On the other hand, some of the docs are great, taking you step by step through what you need. It's uneven. That fellow Gandolf who posts on their forums always says, see the links in my sig--but I don't see any links in his signature.

For real basic usage, it does work more or less out of the box. Of course, we had to do various and sundry things, so I wound up learning a great deal, though
there's still a good deal that escapes me, such as running scripts on remote hosts. The only way I'm able to do it is to create an ssh keypair and have the script generate a single integer which I read on the host machine, then graph that.

[ai-danno]

It's definitely a bear to learn... and you put it right- out of the box it's excellent and does the job. And it can be customized to literally graph anything any way you want it- but the learning curve at that point is, well, extremely steep.

I use it to monitor roughly 50 different routers and switches and the traffic they handle. It's a great program, but it could be better for sure (and I completely agree about the documentation... sigh.)

[scottro]

There was a howto on the forums which cracked me up. The fellow goes into great detail about running a script on remote machines--but all of the detail was about ssh keypairs--basically, he walks you through something relatively easy to figure out, having user cacti being able to ssh to a remote machine.
Then, he ends the howto with something like, create your data inquiry and template, and you're done.
Gee, thanks.

That's a bit of an exaggeration--I think he took you through the data inquiry part. At some point, following their directions, I messed up the total bandwidth template and now can't graph it, but that's not too important. Like you, we use it for routers and switches. We also use it for a few other things--load on a couple of AIX machines and the like. My big accomplishment was getting it to graph the amount of mail in mailq on a machine running postfix. (I used the ssh keypair method, but simply wrote a little script to give an integer of the amount of messages in queue. Then, using some documentation in their handbook that was actually pretty useful, that of getting data from a script, I graph the results of that.

Now, what can be amusing is when you make an opera shortcut to your cacti server's web interface, mistype it slightly, and wind up on a page about growing and cultivating plants.

[ai-danno]

UPDATE- I saw that in 4.3 that rrdtool 1.2 was available and I thought, "maybe this will improve performance"...

... nope. The graphs are, as expected, nicer. But they still generate slowly, and only two at a time.


I'll keep you posted if I find the source of the problem.

[ohauer]

I cannot give any performance data for Cacti at OpenBSD since i'm running Cacti at FreeBSD 7.0.

At the beginning a had mostly problems with mysql (open connections), until i set the wait_timeout parameter to the following value.

file: my.cnf

[mysqld]
max_connections=1000
wait_timeout=900

With the standart Cacti (php) Poller I was not able to process more then 120 DataSources without timeouts ...
A real boost comes from cacti-spine.
With spine I can process now 600+ DataSources and 330+ RRD's in ~5 sec. (network switches + some hosts)

06/22/2008 12:05:01 PM - POLLER: Poller[0] NOTE: Poller Int: '300', 
 Cron Int: '300', Time Since Last: '301', Max Runtime '298', Poller Runs: '1'
06/22/2008 12:05:02 PM - SPINE: Poller[0] Time: 0.9691 s, Threads: 10, Hosts: 5
06/22/2008 12:05:02 PM - SPINE: Poller[0] Time: 1.1113 s, Threads: 10, Hosts: 5
06/22/2008 12:05:03 PM - SYSTEM STATS: Time:2.0844 Method:spine Processes:2 
 Threads:10 Hosts:9 HostsPerProcess:5 DataSources:611 RRDsProcessed:337

[ai-danno]

I'm sorry I didn't respond to this sooner- I really thank you for your insight on this matter... I think at this point that it's related to the front-end graphing issue. It polls and stores data well (pretty much eliminating most of the mysql possibilities.) So I think it's pretty much a graphing/webserving issue at this point.