[plug] What's "too much work at interrupt"?

[James Devenish]

In message <200409071948.51026 at death.2.spammers>
on Tue, Sep 07, 2004 at 07:48:51PM +0800, Bernd Felsche wrote:
> On Tuesday 07 September 2004 14:53, James Devenish wrote:
> > Driver efficiency; hardware tuning?
> I doubt it.
> Athlon 2000 with 512MB of RAM.

So, you're saying that we can assume the CPU is so fast that a bad
driver would be of no consequence? You hadn't previously described the
usage conditions, e.g. did you have 50 users copying files over a
Thinnet network, or 100 users on a switched gigabit link with encrypted
tunnels? Obviously, the fact remains that *something's* generating those
"too much work" Ethernet messages, and they are most often associated
with busy networks => load on the card, and performance issues with the
drivers or interrupts, busyness on the server, etc. The eepro100 driver,
for instance, can flake out completely once it starts generating these
messages. But I have heard that the Intel driver copes better. This
implies there is something in the design of the software or the use of
proprietary hardware features that can circumvent the problem (hence =>
driver efficiency or hardware tuning). To address the "suddenness" of
these errors, can you confirm that you've ruled out abnormalities in the
ambient network activity? (That is, could it be that an external
abnormality is highlighting a latent problem with the software drivers?)