Doesn't add up...

sg · Joined: 01 Jun 2004 Posts: 30

I'm having a problem with my CPU utilization graphs under cacti-0.8.6-pre20040811 ...

They don't add up

See the included graph legend -- Current system util is 21.41, current user util is 60.92, nice util is 0.00 -- total is 26.71?

This is a two-cpu system, the ACTUAL information should reflect that both CPUs are pretty much running flat out at 100% on USER time. I suspect that periodically the system % does go up a bit, but these numbers just don't make sense...

Thoughts?

TheWitness · Posted: Fri Aug 20, 2004 6:29 pm Post subject:

Cactid or cmd.php?

TheWitness

sg · Joined: 01 Jun 2004 Posts: 30

I'm using cmd.php ...

sg · Joined: 01 Jun 2004 Posts: 30

Moved to today's version (cacti-0.8.6-pre20040831) and am still having the same issue.

Interestingly, though, not for every node. 1/8 of my nodes is okay. Examples included.

TheWitness · Posted: Tue Aug 31, 2004 5:08 pm Post subject:

I was just messing with the Graphic Template today and found that unless you add Idle Time to the equation, you will never end up with 100%. Does this help you understand?

I developed a Host Template for HP/UX and Solaris.

TheWitness

sg · Joined: 01 Jun 2004 Posts: 30

Interesting. This may make the graph scale to 100% all the time.

I'm using the stock ucd/net - CPU Usage template from the cacti distrubution.

What I'm really concerned about is that the numbers don't add up. See the "node 2" attachment above -- system is 13.78, user is 94.48, nice is 0.00 and total is 13.13

It could be because the node being monitored has two cpus... but the "node 1" graph is okay -- sytem at 72.65, user at 32.40, nice at 0.00 and total 105.05 -- and that system is exactly the same and has two busy CPUs

hmm.

sg · Joined: 01 Jun 2004 Posts: 30

I downloaded today's snaptshot and - whammo - it seems okay?!

sg · Joined: 01 Jun 2004 Posts: 30

The same problem has remanifested, no changes made by me since the last message.

This is a showstopper....

sg · Joined: 01 Jun 2004 Posts: 30

Been noticing that the problem may be linked to values which add to just over 100%

does this make sense?

TheWitness · Posted: Wed Sep 08, 2004 7:25 pm Post subject:

100% is theoretically the maximum value obtainable. However, due to the fact that the OID's are polled at different times, the CPU utilization can peak over 100%.

Looking at your graphs, I suspect that you have a problem with APPS taking too much CPU. Say for example a race condition of some sort.

Q: are you using the SNMP version or the Perl Script?

TheWitness

sg · Joined: 01 Jun 2004 Posts: 30

Hmm. The attached graph shows that on several occasions the system correctly reports the usage (these are two-cpu systems, so I assume that the maximum CPU utilization can be up to 200%).

Yes, these systems run code that eats 100% of each CPU at times. That's what they're for

What is really bizarre, though, is the addtion. 99.84 + 84.84 m + 0.00 = 113.18 m ?

Thoughts?