100baseT slowness summary

From: Brian Jarvis (jarvis@office-assist.com)
Date: Wed Feb 03 1999 - 11:19:21 CST

The problem I was experiencing with slow 100baseT performance has largely
been cleared up, although not the way I was expecting.

As preported before, I was getting pitiful performance through our 100baseT
network on in-bound network operations to two Sun 250 boxes running Solaris 2.6
connected to a Cisco Catalyst 5000 switch. I could barely get 35 kb/s,
sometimes only 0.5 kb/s, throughput, although outbound operations could
reach into several hundred kb/s.

The advice of almost everyone was to kill autonegotiation of speed and duplex
settings on both the switch and /dev/hme on the Sun boxes, forcing them
manually to do either full or half duplex 100baseT. About half insisted
half duplex was the way to go; the other half quoted Sun documentation
supporting full duplex.

Turning off negotiation and forcing traffic to full duplex improved
inbound throughput to about 250kb/s, a vast improvement but still not
as fast as they should be able to support. Forcing half-duplex on both
ends offered the same level of improvement.

For a lark, I forced the machine to 10baseT, half-duplex. Suddenly,
performance jumps to 800kb/s. I have entirely given up trying to make
any sense of this brain-damaged thing.

Thanks to:
Oleg Olovyannikov
Matt Reynolds
Cyril Liu
Meenakshisundaram Paramasivam
Bismark Espinoza
Benjamin Kearns
"ebs .
Sebastian Benoit
Todd M. Wilkinson
Paul V. Alukal
Ian Parkin
Chris Griffin


Brian A. Jarvis,                     Independent Unix Contractor
jarvis@office-assist.com             9506 Boyer Place, Silver Spring, MD 20910
http://www.office-assist.com/~jarvis tel: (202) 246-0660  fax: (301) 565-4503 

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:13:14 CDT