benchmark

Luigi Rizzo rizzo at icir.org
Fri Jan 5 18:05:38 UTC 2007


On Sat, Jan 06, 2007 at 12:43:50AM +0700, Eugene Grosbein wrote:
> Hi!
> 
> I'm trying to meashure network throughput between two 6.2-PRERELEASE boxes,
> basically get maximim IP packets per second transmitted/received.
> 
> Tried to use iperf from ports in UDP mode with 64 byte payload,
> but it calls gettimeofday() after each write and gives me about 80Kpps only
> for Pentium D 2.8Ghz.

just write your own program that repeatedly calls send() on an udp
socket, and you should be able to go way up.

in the past (2001-2002) i tweaked the kernel with a sysctl that
generated multiple (e.g. 100 or so) copies of each packet for a
single send(), just for testing purposes, and on a 700 MHz machine
i think i reached something in the order of 5-700kpps on a 4.x

At the time the limit was the Gig-E card mounted on a PCI-66/64
bus. These days with a decent card on a PCI-X bus you shouldn't
have these problems.

cheers
luigi

> What alternative should I use? May be, a netgraph node?
> 
> Eugene
> _______________________________________________
> freebsd-stable at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-stable
> To unsubscribe, send any mail to "freebsd-stable-unsubscribe at freebsd.org"


More information about the freebsd-stable mailing list