Slow transfer with 21140 card

[Mark Fardal]

Hi,

> So it's tranmitting at 92 Mbits/s with udp, but only 3.5 with tcp.

some further data for your consideration: setting debug=6 in 
/etc/conf.modules produces a flood of messages when ftp'ing a file
(and a trickle of similar messages otherwise)

Jul 23 01:28:17 weka kernel:  In tulip_rx(), entry 7 05ee0320. 
Jul 23 01:28:17 weka kernel: eth0: interrupt  csr5=0xfc660000 new csr5=0xfc660000. 
Jul 23 01:28:17 weka kernel: eth0: exiting interrupt, csr5=0xfc660000. 
Jul 23 01:28:17 weka kernel: eth0: interrupt  csr5=0xfc670040 new csr5=0xfc660000. 
Jul 23 01:28:17 weka kernel:  In tulip_rx(), entry 8 05ee0320. 
Jul 23 01:28:17 weka kernel: eth0: interrupt  csr5=0xfc660000 new csr5=0xfc660000. 
Jul 23 01:28:17 weka kernel: eth0: exiting interrupt, csr5=0xfc660000. 
Jul 23 01:28:17 weka kernel: eth0: interrupt  csr5=0xfc670040 new csr5=0xfc660000. 
Jul 23 01:28:17 weka kernel:  In tulip_rx(), entry 9 05ee0320. 
Jul 23 01:28:17 weka kernel: eth0: interrupt  csr5=0xfc660000 new csr5=0xfc660000. 

More reasonable output is provided by debug=4:

Jul 23 01:36:08 weka kernel: eth0: exiting interrupt, csr5=0xfc660000. 
Jul 23 01:36:09 weka last message repeated 39 times
Jul 23 01:36:09 weka kernel: eth0: exiting interrupt, csr5=0xfc6e0000. 
Jul 23 01:36:09 weka kernel: eth0: exiting interrupt, csr5=0xfc660000. 
Jul 23 01:36:09 weka last message repeated 172 times
Jul 23 01:36:09 weka kernel: eth0: exiting interrupt, csr5=0xfc6e0000. 
Jul 23 01:36:09 weka kernel: eth0: exiting interrupt, csr5=0xfc660000. 
Jul 23 01:36:09 weka last message repeated 2 times
Jul 23 01:36:09 weka kernel: eth0: exiting interrupt, csr5=0xfc6e0000. 
Jul 23 01:36:09 weka kernel: eth0: exiting interrupt, csr5=0xfc660000. 
Jul 23 01:36:09 weka last message repeated 166 times
Jul 23 01:36:10 weka kernel: eth0: exiting interrupt, csr5=0xfc6e0000. 

So the interrupt message is produced ~400 times a second.

It seems these "exiting interrupt" messages are related to the problem
I'm having.  I tried ttcp with and without the -u (for UDP) flag; the
flood of messages was produced with TCP, but not with UDP.

Perhaps this only restates what you already know, but perhaps it helps
narrow down the problem further.

thanks,
Mark Fardal
UMass