SUMMARY: Sun & Cisco network issue

(I changed the Subject line because it kept getting rejected due to a
"suspicious header" - it was originally "10Mb vs 100Mb")

First of all, thanks to all of the following for their input, suggestions
and info (I received a ton of responses, so I aplogize if I inadvertently
leave someone off):
Stuart Whitby, Paul G. Brol, Bernard du Breuil, Johan Hartzenberg, Alan
Kong, Dennis Martens, Todd M. Wilkinson, Jason Grove, Derek Olsen, Dave
Werth, hmnguyen, Ken Robson, Fran DeMontier, Peter Stokes, Julian Simpson,
Matthew Stier, Jesse Carroll, John F. McLaughlin, Richard Bond, Mark Cain,
Johan Hartzenberg, Vern Walls, Christer Eriksson, Tony Walsh, Joe Fletcher,
Andy Lee, Michael DeSimone, Martin Hepworth, Al Hopper, Steve Holdoway,
Meenakshisun Paramasivam, Simon R, Matthew Ryanczak, Kent Perrier, Ray
McCaffity, Sean P Burke, Matthew Mauzy, Jim Becher, Dik Casper

Short synopsis of problem:  network throughput between two locations was
running faster and more reliably at 10Mb/sec than 100Mb/sec from a Sun E420R
and various Sun workstations (Ultra-1, Ultra-10 & Ultra-60).  Between
systems in the same building, we were getting ~90Mb/sec throughput, but
between buildings ~8Mb/sec.

Troubleshooting:
The majority of the responses I received strongly recommended that we force
both the Sun workstation and the Cisco switch port to 100/fdx.  When we
tried this, we received even more errors and the same throughput - sometimes
even slower.  Sun strongly claims that auto-negotiation works (this specific
problem between Sun & Cisco was fixed about a year ago).  We tried this -
didn't work.  We tried forcing Sun to 100/fdx and the switch to auto - but
the Sun would still drop to 10/hdx.

What we found for certain:
My network admin monitored the switch port-by-port and we started
identifying the systems that were causing the most errors on the switch -
and lo-and-behold, we found they were primarily those that were at 10/hdx
and 2 of these were early models of Ultra-1's.  The port is le0 not hme0 and
"le" can not be configured (from what I've read and Sun told me) - they only
work at 10/hdx.  We moved these 2 systems to the old 3Com switch, hoping to
relieve the collisions and errors on the Cisco switch - nada.

Final result by working with Cisco:  The OS on the switch needed to be
patched/updated to fix a known SNMP memory leak, which caused the majority
of our buffer overflows - thus causing dropped packets and retrans.  By
turning off SNMP on the switch (which we don't need anyway) and using the
"forced" method, we are now achieving 64Mb/sec throughput between the 2
buildings.

Again, thanks to all who replied to my initial cry for help.


Ron Dinwiddie
UNIX System Administrator
Temple Inland Forestry  Products Corp. (TIFPC) - Diboll
email: rdinwid@templeinland.com
Office: 936-829-1592   Fax: 936-829-6666
Pager: 936-630-7086