[netperf-talk] problem with netperf

Rick Jones rick.jones2 at hp.com
Thu Jan 20 10:43:25 PST 2011


[again, please keep the conversation in netperf-talk]

Md Hedayetul Islam wrote:
> Hello Jones,
> 
> Thanks for your mail.If i want to re ./configure --enable-intervals what 
> should i do? should i do both computer i mean both client and server?

The intervals/pacing functionality is strictly on the netperf side, so you don't 
*have* to do both systems, but it is probably a good habit to do both regardless.

I take it this means you tried limiting the socket buffer/window size and that 
didn't help?  Or that using a fixed quantity of data still failed somehow?

happy benchmarking,

rick jones

> 
> thanks
> 
> On Wed, Jan 19, 2011 at 7:33 PM, Rick Jones <rick.jones2 at hp.com 
> <mailto:rick.jones2 at hp.com>> wrote:
> 
>     [Lets keep the discussion on list please as I suspect it may help
>     others later]
> 
> 
>     Md Hedayetul Islam wrote:
> 
>         Hello Thanks for your mail.
>         Actullay I am going to use netperf to measurement data transfer
>         rate from my nerperf client.
>         I have installed netperf server on my ubuntu server and netperf
>         client on ubuntu client.
>         My netperf client is connected with 3G modem(CDMA) and netserver
>         is another publice ip.
>         I can ping my netserver from netclient.
>         when i using UDP_STREAM through netperf  i am getting few
>         info.but when trying to TCP_STREAM getting that error.
> 
> 
>     I'm going to take a wild guess that with the default socket settings
>     and such, the stack allows netperf to queue-up so much data, it has
>     the netserver's test timer expire before the test is actually over.
> 
>     One way to test this hypothesis is to limit the socket buffer sizes
>     with -s and -S options and disable linux's autotuning of the socket
>     buffer/window.  So, something akin to
> 
>     netperf -p 4555 -H 109.123.82.189 -- -m 2048 -s 32K -S 32K
> 
>     or, since this is a TCP_STREAM test, you could control the test by
>     byte count rather than by time and avoid the netserver timer:
> 
>     netperf -p 4555 -H 109.123.82.189 -l -10M -- -m 2048
> 
>     note the minus in front of the 10M - if you omit it, netperf will
>     try to run for 10 megaseconds (M == 1048576).  Use m if you want
>     powers of 10.
> 
>     If that is not sufficient, it may be necessary to re ./configure
>     with --enable-intervals and ask netperf to pace its send() calls.
> 
>     happy benchmarking,
> 
>     rick jones
> 
> 
>         please help as soon as possible
> 
>         Thanks
>         hedayetul
> 
>         On Tue, Jan 18, 2011 at 8:00 PM, Rick Jones <rick.jones2 at hp.com
>         <mailto:rick.jones2 at hp.com> <mailto:rick.jones2 at hp.com
>         <mailto:rick.jones2 at hp.com>>> wrote:
> 
>            Md Hedayetul Islam wrote:
> 
>                I am new in netperf.I just need to data transfer rate from my
>                netperf client to netserver.
>                can u please send me that how to do that?
>                when i use this command got that error.
> 
>                netperf -p 4555 -H 109.123.82.189 -- -m 2048
>                TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
>                109.123.82.189 (109.123.82.189) port 0 AF_INET
>                Interrupted system call
>                netperf: remote error
>                4oxyfi at test-laptopA:/usr/local/bin/netperf-2.4.5$
> 
> 
> 
>            We need to know a great deal more about your situation.  What
>         sort
>            of link joins the two systems?  What sort of systems are the two
>            systems?  Any particular reason you chose to run netserver on a
>            different port number for the control connection? Etc etc...
> 
>            happy benchmarking,
> 
>            rick jones
> 
> 
> 
> 



More information about the netperf-talk mailing list