On Fri, Nov 28, 2014 at 06:45:38AM +1100, Dave Horsfall wrote:
On Thu, 27 Nov 2014, Dave Horsfall wrote:
I've seen a couple of less than flattering
references here; what was the
problem with them?
A number of responses, all of which boil down to the same thing, viz: an
interrupt on each character.
I know that, which is why we (UNSW) re-worked the driver; I can't believe
that we were the only ones who did so. It cleaned out as many silos as it
could, possibly even by disabling interrupts completely and using clock
ticks; I can't remember.
This problem is old as the hills, at SGI we did similar stuff for networking
cards. We could run something like 8 HIPPI cards at full bandwidth with
some 200mhz MIPS cpus. Lots of work in the networking stack to go get the
next packet if it was there after we did this one.