iowait graphics

mtsm · Post by **mtsm** » Wed May 24, 2006 1:43 pm

I tried to modify the existing CPU usage graph template and add iowait ...but the graph became to confuse...i decided to make an IOWait x CPU Idle graph....(separate graph template) it's nice now...thanks for the help!

Post by **gandalf** » Wed May 24, 2006 2:22 pm

Just compiled the modified Graph Template along with ALL xxxCPUxxx OIDs from HOST MIB. Currently, I'm not that satisfied, as it does not show exactly 100 % as it should. See screenshot attached
Reinhard

knebb · Post by **knebb** » Tue Sep 19, 2006 11:31 am

lvm wrote:Just compiled the modified Graph Template along with ALL xxxCPUxxx OIDs from HOST MIB. Currently, I'm not that satisfied, as it does not show exactly 100 % as it should. See screenshot attached
Reinhard

Ok, looks great. Two (three) questions:

1. Can you export this template, so I can download it?

2. Do I really have to touch all existing DataSources?

We have some more hosts....
[EDIT]3. What about SMP-Hosts? Does it graph them right?

knebb · Post by **knebb** » Wed Sep 20, 2006 4:45 am

mtsm wrote:nice!! it's working...i extended ucd/net cpu usage as you suggested

I have to wakeup the old thread- because I have the same problem. I want to monitor the IOWait from my linux-boxes.

I looked in the UCD-SNMP-MIB.txt and found an entry:

Code: Select all

ssCpuRawWait OBJECT-TYPE
    SYNTAX      Counter32
    MAX-ACCESS  read-only
    STATUS      current
    DESCRIPTION
        "iowait CPU time. This is primarily a SysV thingie"
    ::= { systemStats 54 }

So I added a data template copied from the ucd/net - CPU Usage - Nice and changed the related values. As OID I took the "nice" OId (.1.3.6.1.4.1.2021.11.51.0) and changed it to .1.3.6.1.4.1.2021.11.54.0 (54 from the MIB, last line).
Are these steps right?

Ok, then I created a new graph template (copied from 02 - ucd/net - CPU Usage) and inserted four new items (with same setting as the other values like nice and system).
This graph template has been added to a ucd-host.

Now there's my problem, that no IOWaits will be monitored. The rrd-graph is shown and he shows the values, but he shows no iowaits (and there are some!)

In the png at IT-Wait I find "nan" (on System, too

). Why are they "nan"?

Anyone any hints?

Post by **gandalf** » Fri Sep 22, 2006 8:30 am

Please see my signature for a link to NaN Debugging. Post a screenshot and add the full rrdtool graph statement (from Graph Management, DEBUG) to help debugging
Reinhard

knebb · Post by **knebb** » Mon Sep 25, 2006 2:49 am

lvm wrote:Please see my signature for a link to NaN Debugging. Post a screenshot and add the full rrdtool graph statement (from Graph Management, DEBUG) to help debugging
Reinhard

THX for your tips. The nan Links didn't help at all. But I foudn the solution myself:
We used RedHat EL3. Red Hat has patched his 2.4 Kernel to show IO-Waits in top, vmstat and so on.
But they haven't patched their net-snmp packet, which only shows IO-Waits if there is a 2.6 Kernel

I'm on the way to recompile the net-snmp packet to skip the test for the 2.6 kernel.

Instigater · Post by **Instigater** » Tue Sep 26, 2006 4:36 am

lvm, graph exceeds 100% because of system variable is interupt + kernel.

Post by **gandalf** » Sat Sep 30, 2006 4:33 am

Tank you for your inout. Would you prefer just to omit system, then?
Reinhard

cshanerc · Post by **cshanerc** » Fri Jan 26, 2007 10:31 pm

Gandalf,

Would it be possible to post your template? I've created one of my own, but wasn't sure if I had it all right ... so I've been searching and found this thread.

Thanks!

Post by **gandalf** » Sat Jan 27, 2007 11:19 am

Here they are.
Pay Attention! If you are using non-standard rrd definitions, importing this will reset them to default. Please write your current settings down to be able to restore them.
Warning to users of non-standard polling times! The precaution appies to you as well!
Exported using cacti 0.8.6j, fully patched. So you will need at least this version to import

Reinhard

Post by **TheWitness** » Sat Jan 27, 2007 1:15 pm

Only way to fix the 100% is to make it a script and "walk" the OID's so that it get the information from the kernel all in one OS call. By performing multiple snmpget's, you don't achieve this (I think).

This is how it's been with other type of transactions in the past. Problem with this is that it slow's things down of course. Script (outside of script server) slow things down measurably.

TheWitness

Post by **gandalf** » Sat Jan 27, 2007 2:16 pm

TheWitness wrote:Only way to fix the 100% is to make it a script and "walk" the interface ...

Hmm, are we talking about the same topic? I don't get the clue
Reinhard

Post by **TheWitness** » Sat Jan 27, 2007 4:00 pm

I edited my sleepish comment. It's the OID's not the Interfaces Larry (you idiot). Ok enough of that, well I suspect that you poll the entire host cpu information at once, it will pull the information from the operating system all at once thus keeping your total at 100%.

On the other hand, if you make multiple calls similar to the way that the "out-of-the-box" net-snmp graph template does, net-snmp (agent) polls the OS multiple times, resulting in minor deviations from 100%.

TheWitness

cshanerc · Post by **cshanerc** » Sat Jan 27, 2007 5:39 pm

Works great ...

mmuse · Post by **mmuse** » Tue Jan 30, 2007 6:49 pm

I understand there is another thread on multiple CPU (e.g 2-way, 4-way servers). Does this xml takes that in consideration?

Thanks
MM

Cacti

iowait graphics

Template?

Thanks!

Multiple CPU support?

Who is online