[netperf-talk] Extended interim results possible?

[Rick Jones]

Sebastian Clau? wrote:
> Hi,
> 
> As I’ve come through working with netperf I was wondering if there’s any 
> possibility to extend the information given by the interim results every 
> x seconds in demo-mode.
> 
> I.e. is it possible to display the number of packages (with the number 
> of those which were properly received, as well as those which produced 
> errors) in an UDP_STREAM not only at the end of the operation but also 
> every x seconds?

It would be "easy" (SMOP - Small Matter Of Programming) to include the number of 
send calls over the interval, but the number of datagrams properly recieved 
would require the netserver side to send a control message every update 
interval, and that would be a rather larger change.

While it may not work quite right yet, the "omni" tests (./configure 
--enable-omni) should in theory allow a UDP_MAERTS-like test where netperf is 
the receiver.  That would then be able to show actual UDP receive stats in the 
interim results - of course, it wouldn't show send stats since that would be out 
at the netserver.

> As an additional task I’m trying to measure the range of my w-lan by 
> sending a certain amount of packages (with a size of let’s say 1024 
> bytes) per second via a UDP_STREAM to see whether or not they are 
> correctly received by the server. Now I only see the possibility of a 
> bandwidth test, but is there any possibility to also measure the range 
> as described above?

You mean you want to send UDP datagrams at a particular rate yes?  That would be 
./configure --enable-intervals which will then include a global -w option for 
interburst interval, and a -b option for the number of sends in each burst.  The 
interval will be bounded by the granularity of the interval timer on the 
platform on which you are running.  If you add --enable-spin to the ./configure 
options  (and recompile...) then instead of using the interval timer, netperf 
will simply sit and spin between bursts - this burns CPU but gives a much finer 
granularity on the interval.

happy benchmarking,

rick jones