FREEBSD CPU Graphs are wrong

appodictic · Post by **appodictic** » Thu Mar 19, 2009 11:59 am

We have several FREEBSD machines from 6.1,6.2,7.0 etc. None of them seem to have accurate CPU reporting. I am wondering if this is an issue with Kernel Counters, NetSNMP FreeBSD, or cacti issue.

If i look at the top ouptut I see a system that never gets above 20% CPU

last pid: 51704; load averages: 0.28, 0.28, 0.31 up 127+18:21:54 16:45:59
142 processes: 1 running, 140 sleeping, 1 zombie
CPU states: 14.6% user, 0.0% nice, 3.1% system, 0.8% interrupt, 81.5% idle
Mem: 584M Active, 918M Inact, 191M Wired, 115M Cache, 213M Buf, 14M Free
Swap: 1536M Total, 8K Used, 1536M Free

Yet my graphs are constantly showing 40+ percent

I also moved to the advanced templates with IO-WAIT and those numbers make even less sense to me.

Does anyone know how to make sense of freebsd top,ps, and cacti CPU graphs?

Notice how the CPU on web1 totals to 325%?

Post by **TheWitness** » Thu Mar 19, 2009 8:37 pm

Well, until we see the templates, the oid's and so on, we can only guess. My guess is that the first graph is not percentage, but rather time-tics, or maybe a 4 core box.

The sccond graph, well, I don't have a clue. It might be right.

TheWitness

appodictic · Post by **appodictic** » Thu Mar 19, 2009 11:35 pm

Ok i will produce more statistics. Still though for my linux machines these graphs are right on. It is hard for me to fathom that no one has wondered why the CPU graphs for Free BSD do not match top output whatsoever.

appodictic · Post by **appodictic** » Fri Mar 20, 2009 2:42 pm

The template is 'ucd/net - CPU Usage' so the IODs being polled are the standard UCDavis ones. I am using cacti-0.8.7b.

Here is some processor info on a Free BSD7 x86_64 machine
We are using a stock FreeBSD kernel
kern.threads.virtual_cpu: 2
kern.smp.cpus: 2

As you can see my iostat and CPU graphs are not even close. I can find no specific graph templates for FreeBSD. No one on the cacti forums talking about this issue. I am sure people are using cacti to monitor FreeBSD devices. I am noticing this issue across all of our FreeBSD machines 6.1 6.2 7 i386 and x86_64.

But I think this issue would be effecting anyone trying to graph the CPU of a Free BSD system.

Any thoughts?

Post by **TheWitness** » Sun Mar 22, 2009 4:59 pm

Seeing your net-snmp walk of the CPU timetics and percentages along side of the top output would be more beneficial.

TheWitness

appodictic · Post by **appodictic** » Mon Mar 23, 2009 10:44 am

These values look ok:
UCD-SNMP-MIB::ssCpuUser.0 = INTEGER: 15
UCD-SNMP-MIB::ssCpuSystem.0 = INTEGER: 0
UCD-SNMP-MIB::ssCpuIdle.0 = INTEGER: 80
Maybe some calculation involving the raw values is throwing the graph off.

Post by **TheWitness** » Wed Mar 25, 2009 6:41 am

If you are not interestested in I/O wait, you might want to consider these instead for your "new" template.

UCD-SNMP-MIB::ssCpuUser.0 = INTEGER: 15
UCD-SNMP-MIB::ssCpuSystem.0 = INTEGER: 0
UCD-SNMP-MIB::ssCpuIdle.0 = INTEGER: 80

My guess is that the script is using the time-tics, which are problematic to convert to anything like a percentage without a script or script server script.

TheWitness

appodictic · Post by **appodictic** » Wed Mar 25, 2009 10:29 am

Thank you for helping me look at this. Just making sure, You do realize that the script I am using is the default CPU script/template that comes with cacti. This means everyone using this template to graph a Free-BSD system is getting the wrong data.

I am not saying its a cacti problem, FREE-BSD 'isms' like this come up often. I am using the NFSSTAT templates and I had to patch nfsstat to deal with a signed unsigned counter problem.

netdevel · Post by **netdevel** » Thu Apr 23, 2009 7:03 am

appodictic wrote:Thank you for helping me look at this. Just making sure, You do realize that the script I am using is the default CPU script/template that comes with cacti. This means everyone using this template to graph a Free-BSD system is getting the wrong data.

I am not saying its a cacti problem, FREE-BSD 'isms' like this come up often. I am using the NFSSTAT templates and I had to patch nfsstat to deal with a signed unsigned counter problem.

did you solved it?
I'm getting the same problem (freebsd 7.1). My graph shows 125% of cpu usage when the cpu usage is from 95 and 100 %, I don't know where to look to solve this problem

appodictic · Post by **appodictic** » Thu Apr 23, 2009 10:07 am

Unfortunately, it looks like we are going to have to generate our own CPU graphs. Some of the values are correct. Some of them are wrong. (e.g. negative counters etc). I will whip something up in the next couple of days.

I would post a message to the 'Free-BSD users' mailing list, but I think I have been banned because of heated debates about things like this

mscandian · Post by **mscandian** » Tue Aug 03, 2010 12:10 pm

Hi.

I'm getting the same situation. My CPU Usage on FreeBSD 8 (and snmp 5.5) is wrong. The 'top -P' is showing:

CPU 0: 0.8% user, 0.0% nice, 6.0% system, 3.0% interrupt, 90.2% idle
CPU 1: 1.5% user, 0.0% nice, 30.5% system, 0.8% interrupt, 67.3% idle
CPU 2: 0.8% user, 0.0% nice, 0.8% system, 21.4% interrupt, 77.1% idle
CPU 3: 1.1% user, 0.0% nice, 9.0% system, 5.3% interrupt, 84.6% idle

and my graph on cacti is showing System with 98... 99...

What do u did to fix this problem?

Post by **gandalf** » Tue Aug 03, 2010 1:19 pm

net-snmp CPU usage is quite tricky.
First, please know that there are several OIDs related to CPU usage. Not all of them are supported on all systems. For example, there are CPU GAUGEs and at a different OID there are CPU COUNTERs. So it depends on what you are currently monitoring.
I advertise 2nd link of my sig to see what's really going on behind the scene
R.

mscandian · Post by **mscandian** » Tue Aug 03, 2010 2:31 pm

Hi Gandalf.

Thanks for your reply.

In my case, I'm using COUNTER in "Data Source Type" in all templates (ucd/net - CPU Usage - Nice, ucd/net - CPU Usage - System, ucd/net - CPU Usage - User and ucd/net - CPU Usage - Idle).

But the values in 'top' show a low use of cpu with 80%, 90% idle, while cacti shows a high use with 'System'.

Regs

mscandian · Post by **mscandian** » Tue Aug 03, 2010 3:10 pm

I'm trying now using GAUGE but with this, the result is 0 (zero) in all (System, User, Nice and Idle).

GAUGE and OID:

Results with snmpget manually:

lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.50.0
UCD-SNMP-MIB::ssCpuRawUser.0 = Counter32: 889498
lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.51.0
UCD-SNMP-MIB::ssCpuRawNice.0 = Counter32: 426
lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.52.0
UCD-SNMP-MIB::ssCpuRawSystem.0 = Counter32: 20639268
lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.53.0
UCD-SNMP-MIB::ssCpuRawIdle.0 = Counter32: 146133442

Post by **gandalf** » Wed Aug 04, 2010 12:36 pm

mscandian wrote:I'm trying now using GAUGE but with this, the result is 0 (zero) in all (System, User, Nice and Idle).

GAUGE and OID:

Results with snmpget manually:

lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.50.0
UCD-SNMP-MIB::ssCpuRawUser.0 = Counter32: 889498
lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.51.0
UCD-SNMP-MIB::ssCpuRawNice.0 = Counter32: 426
lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.52.0
UCD-SNMP-MIB::ssCpuRawSystem.0 = Counter32: 20639268
lab# snmpget -v 2c -c comu 201.x.y.z 1.3.6.1.4.1.2021.11.53.0
UCD-SNMP-MIB::ssCpuRawIdle.0 = Counter32: 146133442

That are COUNTER OIDs, not GAUGEs
R.

Cacti

FREEBSD CPU Graphs are wrong

FREEBSD CPU Graphs are wrong

Who is online