Hi,
until one hour ago I thought this bug affects only 3com 3c509B cards. Now we
go on switching more and more systems from 2.4.X to 2.6.X and it seems that
many cards work much slower with 2.6.X
with 2.4.27:
=========
bernd@pc-brenn bernd>nttcp -T hamilton2
Bytes Real s CPU s Real-MBit/s CPU-MBit/s Calls Real-C/s CPU-C/s
l 8388608 0.72 0.02 92.7710 3355.4432 2048 2831.15 102400.0
1 8388608 0.73 0.04 92.0581 1864.3941 5783 7932.96 160661.2
with 2.6.11.3:
==========
bernd@pc-brenn bernd>nttcp -T hamilton2
Bytes Real s CPU s Real-MBit/s CPU-MBit/s Calls Real-C/s CPU-C/s
l 8388608 3.29 3.28 20.3957 20.4569 2048 622.43 624.3
1 8388608 3.29 0.04 20.3923 1637.0411 6143 1866.67 149851.2
Cards that show this behaviour:
=======================
Yukon Gigabit Ethernet 10/100/1000Base-T Adapter (sk98lin)
3Com PCI 3c905B Cyclone 100baseTx a
eth0: VIA VT6102 Rhine-II
Cards that work fine with 2.4.X and 2.6.X:
==============================
3Com PCI 3c905C Tornado
The drivers always claim they are using full duplex, here e.g.:
eth0: network connection up using port A
speed: 100
autonegotiation: yes
duplex mode: full
flowctrl: none
irq moderation: disabled
scatter-gather: enabled
I already reported this issue about the 3C509B on LKML several month ago, but
nobody could help. Ever since that mail I got several reports from other
people having the same trouble with their cards.
Initially I wanted to write a bug report following the instructions
in /usr/src/linux/Documentation/networking/vortex.txt, but since this seems
to be more or less a general problem I don't know how to proceed now.
Thanks in advance,
Bernd
--
Bernd Schubert
Physikalisch Chemisches Institut / Theoretische Chemie
Universität Heidelberg
INF 229
69120 Heidelberg
e-mail: bernd.schubert@xxxxxxxxxxxxxxxxxxxxx
|