Michael Heiming <michael+USENET@www.heiming.de> wrote in message news:<492r52-rdu.ln1@news.heiming.de>...
> In comp.os.linux.networking Edu <eduardgv@gmail.com>:
> >> > When I try to measure the performance of a network (e.g. max
> >> > throughput, delay, jitter, loss ratio, etc.). I use ttcp to generate
> >> > (UDP or TCP) traffic, tcpdump to capture and tcptrace to analyze the
> >> > captures. Finally, xplot is needed to print graphs generated by
> >> > tcptrace.
> >> >
> >> > What I need is a better framework, a single tool which provide all the
> >> > functionality.
> [..]
>
> > I've also used mgen or iperf, but yet are necessary supplementary
> > tools for an exhaustive analysis.
>
> Tried netperf?
>
> http://www.netperf.org/

Netperf will give throughput (*_STREAM) and can give round-trip delay
(*_RR). It does not give a direct loss ratio for TCP at least (it
would have to make other calls to retrieve TCP stats - patches always
welcome :), but you can infer one in the UDP_STREAM test taking the
sender versus reeiver stats.

For jitter, you would want to compile -DINTERVALS and then you can
specify -v 2 and see a histogram of either time spent in send()
(*_STREAM) or the round-trip time of each transaction (*_RR).

If you know the typical TCP RTO's for your situation, you could also I
suppose infer a loss ratio from that histogram, taking those points
that were >= RTO as having some number of retransmissions involved.

happy benchmarking,

rick jones

BTW, if you do compile -DINTERVALS, add -DHAVE_GETHRTIME if your
system has gethrtime() - that should help minimize the timing
overhead. netperf 2.3 or later should have that support.