gige benchmark performance

Robert G. Brown rgb at phy.duke.edu
Sat Mar 16 11:12:03 PST 2002


On Thu, 14 Mar 2002, Mark Hartner wrote:

> > also, the latency numbers seemed rather strange to me - 
> > am I understanding the graph correctly, that your eepro100
> > was showing something like 500 us latency?  and that all
> 
> Yes, 541us to be exact.
> 
> > the cards were well above 100 us?  that seems quite high,
> > though I'm comparing to a home-made udp-based ping/echo server.
> 
> Around 120us for both the Intel Pro1000T and Netgear GA620.
> 
> 
> Those are the times it takes to transfer 1byte using TCP/IP with Netpipe.
> 
> Are there better ways to measure latency?

Look at lmbench (www.bitmover.com).  It has tools to measure
"everything".

Although I don't think you'll get a terribly different latency answer
from netpipe and lmbench.  Indeed, if you do a netpipe sweep of packet
sizes, you should see linear growth in bandwidth throughout the latency
dominated region where the slope of the bandwidth/packet size curve is
the number of packets/second.  This gives you a much better "feel" for
the impact of latency at various message sizes than does an isolated
single packet measurement.

   rgb

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu






More information about the Beowulf mailing list