Monitoring CPU stats with SNMP & MRTG or openview

[Wayne Gillan]

Hi all,

I am trying to find a way to get CPU performance data via SNMP and then use MRTG or openview to graph it. I've tried to find MID's that have this but no luck yet. Does anyone know it? All I want is CPU BUSY or something like that.

The server is a HP Proliant RX2660 running V8.3.

If anyone has any ideas that would be much appreciated.

Cheers,
Wayne

[labadie_1]

Hello

I did not know that Vms was available on Proliant :-)

What is Cpu busy ? busy at 100% avec 2 processes in COM on a GS1280, or with 45 processes in COM on an Alpha 1000 ?

It is possible to graph Cpu performance, one example, among many possibilities, hyperspi++

http://pi-net.dyndns.org/hyperspi++

Just click on Cpu and remove anything else.

I do not know how to do this via Snmp, sorry.

[Ian Miller.]

The MIB provided by HP TCPIP are described at

http://h71000.www7.hp.com/doc/82final/6530/6530pro_001.html#mibs_chap

perhaps there is a suitable item you can get

[Steven Fletcher_1 · Accepted solution]

Hi Wayne

We use Cricket/RRDtool for the same job. You need to tell MRTG to query the hrProcessorLoad value (a percentage), 1.3.6.1.2.1.25.3.3.1.2.x, where x is the deviceID as reported by hrDeviceDescr (1.3.6.1.2.1.25.3.2.1.3).

I have had this working on Cricket for the past few weeks but have come across a problem of my own. We have two 4 processor systems in a cluster, however the SNMP process only seems to report on 3 of the processors on both systems:

bash# snmpwalk -v 2c -c 1.3.6.1.2.1.25.3.2.1.3 | grep -i CPU
HOST-RESOURCES-MIB::hrDeviceDescr.1 = STRING: CPU 00 AlphaServer ES45 Model 2
HOST-RESOURCES-MIB::hrDeviceDescr.2 = STRING: CPU 01 AlphaServer ES45 Model 2
HOST-RESOURCES-MIB::hrDeviceDescr.3 = STRING: CPU 02 AlphaServer ES45 Model 2

If anyone can shed any light as to why CPU 03 isn't listed, please let me know!

[Wayne Gillan]

Thanks Steven, those OID's were exactly what I was after. How did you come across them?

It seems though that I have the same problem with our dual processor, CPU 01 won't show.

MOO>snmp_request moo "public" get 1.3.6.1.2.1.25.3.2.1.3.1
1.3.6.1.2.1.25.3.2.1.3.1 = CPU 00 HP rx2660 (1.59GHz/9.0MB)
MOO>snmp_request moo "public" get 1.3.6.1.2.1.25.3.2.1.3.2
- no such name - returned for variable 1

When a walk is performed, it misses the .2 altogether. However it appears to show a correct value for hrProcessorLoad.

MOO>snmp_request moo "public" get 1.3.6.1.2.1.25.3.3.1.2.1
1.3.6.1.2.1.25.3.3.1.2.1 = 2
MOO>snmp_request moo "public" get 1.3.6.1.2.1.25.3.3.1.2.2
1.3.6.1.2.1.25.3.3.1.2.2 = 2

Anyway, I will just have to wait and see how the graph appears after the daily backup.

Thanks again,
Wayne

[Steven Fletcher_1]

Can't remember exactly how I found them, I think I went looking for them on HP's OpenVMS site after the normal CPU OID (that I use for monitoring our Unix machines) didn't work.

Regarding the missing entry for CPU 01 (or CPU 03 in my case), I am willing to bet that over time your CPU 01 processor load will graph exactly the same figures as your CPU 00 processor load.

On our systems, the hrProcessorLoad OID for what should be CPU 03 responds with data but it's an exact copy of the data for CPU 02.... let me know what you find!

[Wayne Gillan]

Yes as you can see from the output above CPU01 (...2) returned the same data as CPU00 (...1).

Thanks again,
Wayne

[Wim Van den Wyngaert]

Not on topic but I start here
http://carsten.familie-doh.de/mibtree/mib-enterprises.html

Wim

[Wim Van den Wyngaert]

This explains the cpu usage
http://www.mnlab.cs.depaul.edu/cgi-bin/sbrowser.cgi?HOST=&OID=HOST-RESOURCES-MIB!hrProcessorLoad (but my test suggest that it's the average of several minutes) and http://www.mnlab.cs.depaul.edu/cgi-bin/sbrowser.cgi?HOST=&OID=HOST-RESOURCES-MIB!hrProcessorLoad&ACTION=WELCOME is a very good walk trhu starting point.

Wim (using snmp to monitor san switches and hub900)

[Wim Van den Wyngaert]

Not on topic but you can minitor the snmp enabled printers on item 1.3.6.1.4.1.11.2.4.3.1.2.0 and it returns "ready to be printed" or "out of paper" or "not connected" etc. Very usefull because VMS is not giving that info in a reliable way.

Wim

[Nandhini]

Hi Am working with SNMP packets. By Passing this 1.3.6.1.4.1.11.2.4.3.1.2.0 OID am able to get the status of the printer. But my problem is

Am broadcasting the SNMP packets over my network. So all the printers will respond. I need to get this status along with the IP address of that printer.

I dont know how to send data packets with more than one variable bindings.

[Nandhini]

Hi Am working with SNMP packets. By Passing this 1.3.6.1.4.1.11.2.4.3.1.2.0 OID am able to get the status of the printer. But my problem is

Am broadcasting the SNMP packets over my network. So all the printers will respond. I need to get this status along with the IP address of that printer.

I dont know how to send data packets with more than one variable bindings.

[John Gillings]

Wayne,

Have you tried using T4 instead? You can find T4 at the OpenVMS web site or at trendsthatmatter.com . Install T4 on your OpenVMS system, then look at the data using TLViz on a PC. It will give you very nice timeline graphs of CPU BUSY, or CPU in various modes, or lost of other things. It will also allow you to correlate CPU with other measurements, which goes a long way to answering "OK, so the CPU's busy, but what's causing it?".

[Wayne Gillan]

John,

Yes we are already using T4 with TLViz as well. But we just wanted something which gave a more general overview of how the system is doing, say compared to 6 months or a year ago. T4 captures every bit of information you could possibly need, but it's not that quick and easy to view. You have to know which day or days you wanted, then get the files across to your pc (we tried samba/cifs for this) and then open the files with TLViz. Compared to MRTG where you can view the graphs imediately from anywhere you have access to a web browser. We'll continue to use T4 however, just in case we need to look at a particular days stats in detail.

Cheers,
Wayne

[Wim Van den Wyngaert]

Nandhini,

How are you broadcasting your stuff ?
The IP can be found in 1.3.6.1.4.1.11.2.4.3.5.2.0

Wim

[Guinaudeau]

Steven wrote :

"Regarding the missing entry for CPU 01 (or CPU 03 in my case), I am willing to bet that over time your CPU 01 processor load will graph exactly the same figures as your CPU 00 processor load. ..."

Steven, Greater :

i found and used this thread first to retrieve under under V7.3-2 TCPIP V5.4 platform the hrProcessorLoad OID and i got a reasonable CPU load result (at least under DS10 test machine, cannot say about 2- or 4-CPUs machine and that release).

i estimated and did not measure but it looked like the value was not averaged on the
last minute as RFC1514 indicated, there might be a slight difference somewhere or i did a mistake. nevertheless something reasonable.

now i tried to get the same result with

OpenVMS V8.3 UPDATE V7.0
TCPIP V5.6-9ECO2

first under a DS10 machine :

$ snmp_request localhost "public" get 1.3.6.1.2.1.25.3.3.1.2.1

=> it simply returns ever 0 whatever the CPU load !!!

on a AlphaServer ES45 with 2 CPUs : it behaves like you predicted and/or observed.

questions :

=> did someone already install the brand new TCPIP V5.6 ECO3 and could say about its behaviour with that snmp_request ?

=> did someone observe also such abnormal behaviour with other hr objects ?

=> possibly since TCPIP component LBROKER depends upon CPU load too : did someone observe abnormal metricview results ?

[Guinaudeau]

John wrote :

"Have you tried using T4 instead?"

Greater wrote :

"Yes we are already using T4 with TLViz as well. But we just wanted something which gave a more general overview of how the system is doing, say compared to 6 months or a year ago. T4 captures every bit
of infos ..."

i agree "yes and no" with Greater that T4 [or TDC or other] collector(s) and bench of tools may not be the right answer at that place.

To be precise :

1)

You can use postprocessors of T4 collection tools to compress / archive the collection, so to observe long term seasonal or yearly
variations :

-> at least CSVPNG is a powerfull tool to get a small CSV (selecting only few params, averaging on a longer sampling period, ...)

"CSVPNG" name may confuse seom people : you will find CSV->PNG generator features, but also a lot of CSV->CSV features in that tool

-> VMS bootcamp attendees may know about other usefull "T4 archivers" ?

2)

I would name CPU load value a "quasi-instant" value = an average on the very recent past (last minute or last seconds in LBROKER metrics).

To retrieve it from T4 collections, you may need to "cut" the collection file to analyse the past and retrieve that value on the
last period of concern => this is not convenient