Netperf places no restrictions on the size of the send or the receive
socket buffers. It will depend entirely on how large the Transport lets
the sockets be, and/or how much RAM can be malloc'ed
> 2.Does 'Send Message Size' mean 'one packet size'?
Not necessarily. One should only assume that "Send Message Size" is the
quantity of data presented to the Transport in each send() call. Whether
or not that is the packet size depends on a whole host of other factors.
> 3.Can I benchmark server tcp throughput such as below
> topology?
> Ethernet +----------+
> +----------+ Client +
> | +----------+
> |
> |
> +++++++++++ | Myrinet +----------+
> + Server + ----+----------+ Client +
> +++++++++++ +----------+
>
> I would like to say that 'Can I benchmark above topology that
> Server send to multiple client over different physical layer?'
You can, if you invoke netperf multiple times in parallel. As there is
no synchronization in netperf 2, it is probably best to use either
really long run times, or to have three tests in sequence to each
client, both clients in parallel, and then only take the values of the
middle (second) data point. That way you have a decent assurance that
both clients were active at the same time.
rick jones
-- these opinions are mine, all mine; HP might not want them anyway... :) feel free to email, or post, but please do not do both... my email address is raj in the cup.hp.com domain...