[netperf-talk] netperf simultaneous threads question

[Rick Jones]

mi wake wrote:
> 2011/1/20 Rick Jones <rick.jones2 at hp.com>:
> 
>>mi wake wrote:
>>
>>>I get netperf4,but still can not find config xml file example about
>>>using simultaneous threads.
>>>How can i run 300 instances of netperf TCP_RR test on one pc？
>>
>>In the context of netperf4, which may or may not be in a "working" state, it
>>would involve adding 600 separate "test" declarations to the config file,
>>300 for the receive side, and 300 for the send side. And some number of
>>netserver declarations, depending on your test setup.
>>
>>Suffice it to say, as it stands (or falls-over) today, the UI for netperf4
>>is not overly friendly.
>>
>>Lets back-up a bit - to what end are you trying to have 300 instances of
>>netperf going?
> 
> 
> I  use that to test the linux new features RPS(Receive packet steering).
> Tom give us Benchmarks of 500 instances of netperf TCP_RR test.
> I am curious about that result,so I want to test that in the same way
> to see what happens.

If you are willing to get the throughput via non-netperf means, then I would 
suggest simply starting long-running netperf2 TCP_RR tests via scripts, perhaps 
with for loops per the manual: 
http://www.netperf.org/svn/netperf2/trunk/doc/netperf.html#Using-Netperf-to-Measure-Aggregate-Performance

You could, I suppose, try to launch all 500 netperf TCP_RR's at once (from a 
loop, targetting many remote systems) and set a long run time and ask for a full 
30 iteration set of confidence intervals and get numbers directly from netperf, 
but I think for this scale it is best to pull the stats out of band.

happy benchmarking,

rick jones