SUMMARY: 10Mbp/s switches

From: Ken Ferguson (
Date: Tue Oct 22 1996 - 11:55:06 CDT

My Original Question:

> <> I have just installed some new 10Mbp/s switches from cabletron for the Unix
> <> workstations here. I did some testing after moving a small part of the sun
> <> network to the new switch. I essentially ran a script that would run the
> <> command
> <>
> <> /bin/time rcp file target:/dev/null
> <>
> <> where file were files that were 1, 5 10, & 20MB and target was multiple
> <> machines both on the new switched network and on the old shared network. I
> <> ran each 4 times threw away the first time and averaged the remaining 3,
> <> then divided the file
> <> size by the time to get throughput. The best time I got was 880kB/s, I was
> <> expecting 1250kB/s (10Mb / 8b per byte = 1.25MB or 1250kB). Their support
> <> has now told me that that is about the best I can expect given the machines
> <> ( they are mostly sparc5's) the idea being that the computer cannot drive
> <> the le0 at 10Mbps. They have tests supporting the 65-70% throughput number
> <> (still better than competitors).
> <>

I would like to thank everyone for the help. (John Justin Hough)
james mularadelis <>
Reto Lichtensteiger <>
Casper Dik <casper@holland.Sun.COM>
Tom Southerland <> (Russell David) (Ron Loftin)
Tom Mornini <> (Richard Read)
Michael Maciolek <>

The overall consensus was that 800k/sec was a reasonable number to expect from a switched 10Mbps network. My original expectation of 1250Kb/s was wrong, there is an overhead. Michael Maciolek sent me a very nice copy of an explanation of the overhead and that number comes to just under 10%. So the very best you could expect would be 1125kb/sec assuming everything else was perfect. Everyone seemed to think that the ethernet port could easily be driven at 10Mbp/s. I also got some ideas of different test to test with. Using spray I was able to approach 1100kb/s. Utilizing multiples of this I was able to prove the switched 10 continued to provide 1100kb/s under multiple sessions on the same card, while the shared 10 performance degraded as you would expect under multiple sessions down to about 400kb/s.

So it looks like the rest of the network now gets migrated to the switched 10.


This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:13 CDT