[netperf-talk] exact behaviour of generated burst traffic

[Rick Jones]

Kristijan Dragicevic wrote:
> Hello,
> 
> I am trying to generate bursty traffic with a command like
> netperf -t UDP_STREAM -H 10.0.2.2 -l 10 -- -m 228 -w 100 -b 1024
> but I do not see a difference to the behaviour of using command
> netperf -t UDP_STREAM -H 10.0.2.2 -l 10 -- -m 228
> no matter what parameters I use for the -w and -b option


Just to be pedantic, you did configure netperf with burst mode enabled yes?

> I browsed the whole web to find some kind of specification for these
> options. Browsing the code just lets me know, that the packet size of
> the burst packets seem to be the same as the packet size of the
> ordinary emitted packets.

Yes.

> So my question is:
> What happens during the burst intervals compared to the time between
> the intervals? Are there different packets sent out or in another rate?
> Well, I know that -w specifies the interval in ms, and -b specifies
> the amount of packets to be emitted. But as far as I can see it just
> looks like netperf is emitting packets "like there is know tomorrow" all
> the time.

-w sets an interval timer, so it is how often a burst should be sent, 
not an interval between bursts.  the expectation is that netperf 
finishes the -b burst of sends before that interval timer pops, and sits 
in a sigsuspend() waiting for the interval timer (this assumes that one 
isn't using --enable-spin from top of trunk at 
http://www.netperf.org/svn/netperf2/trunk/)

If it takes longer than -w to send the -b burst, it can appear that 
netperf is not bursting at all.  It can also appear that netperf will 
finish "early" on some systems that do not have (or netperf does not 
know has) siginfo support to be able to restart selected system calls 
based on what the signal handler does.

hth,

rick jones