[netperf-talk] netperf in windows environment

Mon Sep 18 06:08:33 PDT 2006

Rick,

  I tried to run the test with -v 3 option. I am running my tests on MP system with processors. I used -c and -C both the options. It gives the CPU Utilization and interrupt/sec in three columns. ( I am off today, tomorrow I will cut and paste the result here.)

  The problem what I have it gives output something (IIRC) like following:

  Total CPU Utilization   CPU-1 Utilization        CPU-2 Utilization

  Total Int/sec                Int/sec (for CPU-1)     Int/sec (for CPU-2)

  But when I tried to do the math like:

  Total CPU = CPU-1 + CPU-2 

  This does not seem to match. Rather IIRC, "Total CPU" was somehow ;less than "CPU-1". I had the similar observation for the "interrupts/second" metric also.

  I will try to add the histogram to see the individual round trip.

  For the problem what I see while trying to run the "netserver" on the Dell 2900 system, I am not sure what is going on there. The same code base seems to be working on the different Windows system. I am not sure if anyone elase has seen this issue.

  If I got the Service Demand explaination correctly, it is kind of normalization of 
  "Service Demand = CPU Utilization / Amont of Data Transfer". Am I correct? I will go thru the documentation to get more ifo about this. 

  Thanks for the help.

Rick Jones <rick.jones2 at hp.com> wrote:
  For the sake of posterity, let's keep this in the netperf-talk list. It 
may help others in the future.

anil mishra wrote:
> Rick,
> 
> Thanks for the reply.
> 
> I am using Netperf 2.4 + downloaded from the trunk.
> I am not sure what is IIRC in your reply. I tried to use the -c option 
> to get the CPU utilization while running my test, it gives me following :

IIRC == If I Recall Correctly

> TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.14.52.155 
> (10.14.52.155) port 0 AF_INET
> Recv Send Send Utilization Service 
> Demand
> Socket Socket Message Elapsed Send Recv Send Recv
> Size Size Size Time Throughput local remote local 
> remote
> bytes bytes bytes secs. 10^6bits/s % N % U us/KB us/KB
> 8192 8192 8192 10.00 1068.93 92.89 -1.00 14.238 
> -1.000
> 
> I am not sure if CPU Utilization (92.8 for local and -1.00 for remote ) 
> values are correct.

Since -c only requests local CPU utilization, the -1.00 for remote is 
indeed correct. If you want to also see remote CPU utilization you need 
to add -C to the netperf global command options.

If your's is a UP system, depending on the CPU horsepower I could 
believe that ~93% CPU util was correct.

> What is service demand metric printed in the result. 

IIRC :) described in the manual :) Normalization of throughput and CPU 
utilization, showing how much CPU resource was consumed per KB of data 
transferred. Lower is better.

> 
> Can we also get the # of interrupts/sec metric from the Netperf?
> 

Nope.

> 
> Yes, good point, I need to convert the script to .bat. Can I use them as 
> it is in the Cygwin environment?

Perhaps, I've never tried it before.

> 
> Yes, you have done your math correctly.
> 
> I got the following for the # trans/sec.
> 
> TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 
> 192.168.1.100
> (192.168.1.100) port 0 AF_INET
> Local /Remote
> Socket Size Request Resp. Elapsed Trans.
> Send Recv Size Size Time Rate
> bytes Bytes bytes bytes secs. per sec
> 8192 8192 1 1 10.00 17150.77
> 8192 8192

Then indeed, the average round-trip latency would be 1/17150.11 seconds. 
You could probably ass-u-me that the one-way latency was that over 
two. If you want to see individual rount-trip times, you need to enable 
the histogram support in the stuff in NetPerfDir - adding it to the 
config.h that is there (IIRC) or adding it as a -D

> I am trying to run the netserver on the Dell Poweredge 2900 system and 
> it gives me following error messages:
> 
> C:\netperf>netserver
> netserver: fopen of debug file as new stdout failed!: The system cannot 
> find the path specifie
> 
> Am I missing something very obvious?

Likely as not, the code is it may be ass-u-me-ing something about paths 
that is Unix-like specific. Somewhere in src/netserver.c I would think.

> Thanks much in advance for your help and for such a nice benchmarking tool.

my, and I suspect all those who have contributed, pleasure,

rick

> R
> 
> 
> 
> */Rick Jones /* wrote:
> 
> anil mishra wrote:
> > Hi,
> > I am trying to run the netperf in the Windows 2003 environment. Can
> > someone let me know how to get the %CPU Utilization in the Windows
> > environment.
> 
> Which version of netperf are you using? I would suggest at least 2.3 if
> not 2.4 from top of trunk in the subversion repository:
> 
> http://www.netperf.org/svn/netperf2/trunk
> 
> > It looks like the configure script and Makefile has to be changed
> to get
> > the CPU Utilization.
> 
> Netperf doesn't use configure under Windows. The README.windows file
> should describe how netperf is compiled under windows. IIRC that will
> include the Windows-specific CPU utilization measurement code you can
> access via -c/-C just like the rest of the platforms.
> 
> > Has anyone done in it Windows environment?
> >
> > What do I need to do run all the available scripts in the Windows
> > environment?
> 
> The set of "all available scripts" for netperf under Windows is the
> empty set as none of them have been ported to bat files. So, the first
> thing you would need to do is port them over to bat files.
> 
> > I just tried to run the Netperf on Brodcom NIC and got the
> latency as ~
> > 58 usec. Does it seem to be correct?
> 
> If I've done my math correctly, that suggests you are seeing ~17K
> Transactions/second yes? Or is that one-way latency you are reporting?
> 
> rick jones
> 
> 
> ------------------------------------------------------------------------
> Talk is cheap. Use Yahoo! Messenger to make PC-to-Phone calls. Great 
> rates starting at 1¢/min. 
> 

---------------------------------
Stay in the know. Pulse on the new Yahoo.com.  Check it out. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.netperf.org/pipermail/netperf-talk/attachments/20060918/01ad2c4e/attachment.htm