On Thu, 12 Aug 1999 15:11:53 -0400 
 "Michael T. Stolarchuk" <mts@off.to> wrote:

 > it has been EXACTLY like slow interrupt response time.  That was one
 > of the belief's i've had.  It may also explains why linux doesn't have
 > the same problem; they keep advertising how hard they struggle to keep
 > interrupt performance very low.

Interrupt performance very low?  :-)

Anyhow, I suspect the problem isn't actually "slow interrupt response time",
at least on the part of the NetBSD kernel.  Especially in the case of
100mbps Ethernet, the kernel can certainly dispatch the interrupt handler
in plenty of time, and even if it doesn't, the way the PCI network drivers
are constructed, this doesn't matter a whole lot (unless it's not dispatched
for a LONG time, say, 64 frames worth).

        -- Jason R. Thorpe <thorpej@nas.nasa.gov>