[netperf-talk] Some question about CPU utilization

[Benjamin Thery]

Rick Jones wrote:
> Benjamin Thery wrote:
>> Hi,
>>
>> I'm using netperf to do some network benchmarking on Linux and
>> I have some questions regarding CPU utilization.
>>
>> Very often, um... most of the time, when I run netperf I get the
>> "!!! WARNING, Desired confidence was not achieved within the specified
>> iterations.". It's _always_ due to variations in CPU util.
> 
> How far away from your desired interval (10%) is netperf reporting?

A lot. Most of the times it is between 15% and 30%. Sometimes it is
50-60%, I even got a 365% once (but CPU util was only 0.19% this time).

> 
>> What can I do to stabilize the CPU util on my host?
>> Are there some known tricks?
> 
> One is to bind netperf and netserver to a specific CPU with the global 
> -T option.

Ok, I will try that.

>> Should I increase the number of iterations or the duration of the test?
>>
>> Here are some additional informations:
>>
>> I run netperf with the following options : -I 95,10 -i 10,2
>> and I set the test duration to 40 seconds.
> 
> I'd suggest going with -i 30,3 .  30 is the maximum netperf will do, and 
> 3 is the minimum - your setting was getting silently set to 3 for the 
> minimum :)

>> CPU util is measured between 1 and 2% (very low CPU load).
>>
>> My test machine is a bi-processor dual-core Xeon 5130 2GHz.
> 
> What sort of test is this with that low a CPU load?  IIRC the linux 
> procstat CPU util method is "statistical" and at such a low load level I 
> get a little "concerned" about the measurments. 

I run the following tests: TCP_STREAM, TCP_MAERTS, TCP_RR, UDP_STREAM,
UDP_RR. netperf never reports a CPU load greater than 3% with these
tests on my config. I'll attach one of my test result at the end of this
message, if you want to have a look.

I also understand that with such low CPU load it is easy to have
variation in tens of percent (eg. if another process wakes up on the
system).

> If going to 30 
> iterations doesn't give you stability, you may also want to try 
> increasing the iteration runtime to 120 seconds.

Yesterday, I increased iteration time to 120 seconds. It was better but
not perfect. I'll follow your advice and increase iterations today.

>> I tried to run CPU_LOC test.
>> It returns immediatly and givethe value 100.
>> (I thought I read in the manual this test should last 40 sec?)
> 
> You may have - although depending on which manual you are reading, it 
> may also discuss now how some CPU util methods don't require 
> calibration, so the *_CPU tests will return "immediately."  The linux 
> procstat method is such a method.

I must have skipped this part of the manual :)

Thanks for your advices.

Benjamin

PS: See netperf output below.

> 
> happy benchmarking,
> 
> rick jones
> 
> 


-- 
B e n j a m i n   T h e r y  - BULL/DT/Open Software R&D

    http://www.bull.com
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: netperf-results-nptl-1400-2.6.25-rc1-nonetns-nosysfs-rootnetns-physitf.txt
Url: http://www.netperf.org/pipermail/netperf-talk/attachments/20080220/2e9631fe/attachment.txt