[vortex] Poor performance with 3c595

[Donald Becker]

On Thu, 12 Jul 2001, Simon Derr wrote:

> I'm trying to use 3c595-TX cards on our new PIII-1GhZ, and i am really
> disappointed by the performances :

I though I misread that line until you posted the confirmation.
A PIII combined with a 3c595?

> 3c59x.c 18Feb01 Donald Becker and others
> http://www.scyld.com/network/vortex.html
> eth1: 3Com 3c595 Vortex 100baseTx at 0x5440,  00:a0:24:9d:49:93, IRQ 9
>   64K word-wide RAM 3:1 Rx:Tx split, autoselect/10baseT interface.

Keep in mind that the 3c595 was notable as one of the very first
100baseTx PCI cards.  It used a pretty basic interface.


I am surpised at the low performance you are getting.  The 3c59x should
be able to do about 80Mbps using 50% of a P5-100.  I'm guessing that the
driver version you are using has aged badly -- perhaps some change
results in the driver doing a spin-wait where it used to be interrupt
driven.

> I also tried to set options=0x214 to force Bus-master (as shown in
> vortex.txt) but then I get lots and lots of :
> 
> eth1: Too much work in interrupt, status e101.

Yup, something changed.

But to be realistic, the 3c595 is ancient.
  It is thrashed in performance by an $8 rtl8139 board
  It doesn't have autonegotiation,
  There is no safe way to detect the link speed
  There are various errata when actually using bus mastering on Rx

Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993