gaps in graphs and is it possible to graph NUMA system?

Snake007uk · Post by **Snake007uk** » Fri Feb 23, 2007 4:22 am

Hi,

I am having this problem with my cpu graphs, i have attached two pitcures to make it easier to describe the problem

Basicaly I a getting gaps in my cpu graphs for user cpu activity, this is only happening to a specific group of servers (all in the same cluster). These nodes are in a busy-ness loadbalancing.

Its funny as some on the gaps are 20mins long? but the system cpu graph is still going?

The other thing is we are also using MRTG (are planning to replace with Cacti), MRTG doesnt show these drops?

the cacti log file for that particular time doesnt show anything, I am going to turn on debugging to see if what it does show for that time period, as we noticed the issue at that time.

Snake007uk · Post by **Snake007uk** » Fri Feb 23, 2007 5:32 am

sorry I also wanted knqo id it was possible to graph NUMA systems

Post by **gandalf** » Fri Feb 23, 2007 3:51 pm

Snake007uk wrote:sorry I also wanted knqo id it was possible to graph NUMA systems

FInd some help for this at the first link of my signature. But be prepared not to find a complete solution. It's will help you building your own scripts
Reinhard

Snake007uk · Post by **Snake007uk** » Mon Feb 26, 2007 6:21 am

Thanks gandalf,

I was thinking it maybe that, there may not be enough threads for cacti to produce the graphs in time.

Snake007uk · Post by **Snake007uk** » Mon Feb 26, 2007 7:49 am

Looking into this a bit more, I dont understand why at some points there are gaps and others there are not? and the funny thing is it only happens to a certain cluster of server graphs no other cluster (tree) has the problem?

Snake007uk · Post by **Snake007uk** » Mon Feb 26, 2007 12:38 pm

below is an output of my poller pref, anyone have any ideas?

Post by **gandalf** » Fri Mar 02, 2007 2:45 pm

Well, this one is handled at the second link of my signature. Intermitting problems are a bit harded to track, but try the debug hints first
Reinhard

Snake007uk · Post by **Snake007uk** » Mon Mar 05, 2007 5:22 pm

I think I solved my problem doing three things

1) Increase the graph Y height to 200 as I belive the CPU's were going over 100% (dual xeon cpu's)
2) RDDtool tune to make sure all rrd templates where updates...
3) I turned off RIGID RRD setting for the CPU graph.

I hope this helps

Post by **gandalf** » Tue Mar 06, 2007 2:35 pm

Yeah, good! The multiprocessor issue; i missed it. But this is definitively the correct approach!
Reinhard

Snake007uk · Post by **Snake007uk** » Tue Mar 06, 2007 3:48 pm

I thought multi processor graphs would grow 100% * number of cpu?

Post by **gandalf** » Tue Mar 06, 2007 3:52 pm

Yep. But pay attention: There was some design change with latest (AKA 5.4) version of snmp. I do not remember exactly: it was either "system load" that has changed or processor cpu
Reinhard

gaps in graphs and is it possible to graph NUMA system?

gaps in graphs and is it possible to graph NUMA system?

Who is online