--- Christian Lyra <[EMAIL PROTECTED]> wrote:
> How this compares to "standart" throughoput test? like one with "nc -l
> <port> > /dev/null" on one side and "cat /dev/zero | nc <ip>
> <port>" on the other? my guess is that UML itself will be the
> bottleneck.
I think you are right ... mostly.
Without a preload, ping is very slow due to the context switching required
(50ms round-trip for each packet) - so you're right there.
With preload (the test I performed) I can increase the throughput until the
system saturates on CPU. The UML host was at about 80% CPU during this test -
50% user, 30% system across 2 CPUs. So in this case I think it is not
constrained too much by UML and context switching. I have already noted a lot
of code points to optimise, but I'm not ready to focus on optimisation until
more functionally complete.
--
Rich
Get the name you always wanted with the new y7mail email address.
www.yahoo7.com.au/y7mail
_______________________________________________
Ntop-dev mailing list
[email protected]
http://listgateway.unipi.it/mailman/listinfo/ntop-dev
