poller time follows disk i/o

[wjm]

I have setup for 1 min polling, and sometimes it would keep going over one minute.
Then it would mysteriously go away.
Other that yum updates, this box is not messed around with.
Has anyone else seen this relationship to disk writes?

I am running a CactiEZ install with 0.8.6h. (thanks jimmy)
The hareware is an IBM blade 2 Xenon 4 gig ram, writing to raid 1 scsi drive

Time:33.4140 Method:cactid Processes:4 Threads:60 Hosts:278 HostsPerProcess:70 DataSources:12485 RRDsProcessed:5291

[/img]

[gandalf]

Interesting find and nice graphs. But it is not surprising. Cacti polling time largely depends on rrdtool update performance which in trun is closely related to disk write speed.
So, the question that is left is: why is disk performance changing?
Reinhard

[wjm]

I have been using this server since the summer.
When the polling issue first happened (I was not polling local disks), I started yelling at the guys that manage the blade center, then it went away.
It came back again , then I thought I had too many rrds. It wasn't until I was quickly scrolling through the monthly graphs that I saw a pattern.
The tan colour is disk writes, what could make a two week cycle?
This server has been rebooted numerous times within the time span with no affect (effect?) on the polling or disk writes.

[gandalf]

I have been using this server since the summer.
When the polling issue first happened (I was not polling local disks), I started yelling at the guys that manage the blade center, then it went away.

So start yelling again
To be honest, I've never heard of such a behaviour
Reinhard

[wjm]

My lunch is starting to wear off, so I will wait until Monday.
I am thinking of doing a clean rebuild of 0.8.6j with a raid 0 to squeeze a little more out of the hard drives. If I can b*tch into getting space on the fibre channel SAN, could that be a better place to put the mysql database?
The more people that see (understand) the value, the more switches and servers get added.

I commend your dedication and support, mrtg and now cacti has been such a useful tool to explain to users how their network and servers are behaving (or mis-behaving)

[gandalf]

Best practise is to put rrd files on the fastest disks (write speed matters). Until now, I've not heard of issues related to mysql performance. But you may google the forums for some (few) hints on tuning mysql. Apart from mysql connections, I'm sticking to standard RHEL settings.
But if there are other applications running in parallel, things may change
Reinhard