21143 problem

[Robert G. Brown]

On Wed, 4 Nov 1998, Douglas Eadline wrote:

> netperf calls:
> ./netperf -t UDP_STREAM -p 12865 -n 2 -l 60 -H coyote3 -- -s 65535 -m 1472
> ./netperf -t TCP_STREAM -p 12865 -n 2 -l 60 -H coyote3
> 
> The netperf results are:
> 
> UDP UNIDIRECTIONAL SEND TEST to coyote3
> Socket  Message  Elapsed      Messages
> Size    Size     Time         Okay Errors   Throughput
> bytes   bytes    secs            #      #   10^6bits/sec
> 
>  65535    1472   60.00      487486      0      95.68
>  65535           60.00      487458             95.68
> 
> TCP STREAM TEST to coyote3
> Recv   Send    Send
> Socket Socket  Message  Elapsed
> Size   Size    Size     Time     Throughput
> bytes  bytes   bytes    secs.    10^6bits/sec
> 
>  65535  65535  65535    60.04       3.65

This is very similar to a problem I reported some months ago for a Cisco
Cat5000 switch.  We found empirically that if one set the port speed of
the switch externally (e.g., actually set the port to 100BT FD with the
switch control software) that this would occur.  Some sort of framing
error was occuring that allowed packets through but corrupted them.  I'm
sure that if one actually examined the UDP packets received that there
are bit errors in most of them.  We saw this with more than one kind of
NIC but it was worst with tulips.

When we set the ports to auto (so they would negotiate with the NIC) the
problem went away.  The only person at the time who reported seeing the
same thing was Greg Lindahl on pretty much the same hardware.  This
sounds like the second time it might have occurred.  Is the switch port
set to autonegotiate or did you set it to 100/FD explicitly?  It may be
that some cards and some switches don't negotiate correctly unless
permitted to negotiate all the way.

   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@phy.duke.edu