> On Sun, 24 Mar 1996, Chris Evans wrote:
>
> > Hope you can get the 3c590 driver working soon. Incidentally, did the
> > fixing patch posted here a few days ago work? And without my loss of my
> > 800k+/sec performance?
>
> I measured the performance of my revised version of the "Vortex" driver.
> I have two 3c590 cards, one with 8K buffer and one with 32K buffer. In the
> former case I had 510 KB/sec, in the latter case I had 650 KB/sec. With
> the old "Vortex" driver I had nothing significant in the former case, since
> it didn't work, and 650 KB/sec in the latter case.
>
> I think that if you have a 3c590 equipped with a buffer of 32KB, my patch
> won't hurt your performance. If you have a buffer of 8K, the loss of
^^^^^^^^^^^^^^^^^^^^^^^^^^
> performance will be significant.
>
I have an 8k buffer. Why should buffersize make such a drastic
performance, assuming a server that isn't loaded? Or is the driver
_really_ broken, as Alan suggested in a "things to do" e-mail.
If the patch leaves receiving as it was (ie quick,
problemless), and reduces transmission performance without leaving the
CPU in busy-wait for the extra delay (ie by executing a delay loop within
an interrupt, with other ints disabled), I'll try the patch out (please
post it to me personally)
That's if I'm around much in the next few weeks, which is unfortunately
unlikely. However, knowing Linux development, when I return, the problem
will have been ironed out perfectly :-)
Cheers,
-- Chris