Re: Macintosh kernel out

David S. Miller (davem@caip.rutgers.edu)
Tue, 21 May 1996 13:01:35 -0400


Date: Tue, 21 May 1996 12:48:49 -0400 (EDT)
From: Matthew Kramer <kram1968@nova.gmi.edu>

> Not much I guess. You can simplify lots of the driver functionality I
> think, so that you can reduce most of the overhead to the overhead of
> function calls.

Your programs still make the calls to the linux kernel, which has to
identify the calls, then make the approiate call to the microkernel. If
a return value is given, it is sent back to the kernel, which sends it to
the program. Too much overhead.

Agreed, I'm to the point with the Sparc port where I'm counting cache
lines and pipeline stalls. When a microkernel based version of Linux
can get close to 7mb/s localhost TCP bandwidth on a 40mhz processor
like I currently am, then I would start considering a microkernel as
a viable thing to even use.

Later,
David S. Miller
davem@caip.rutgers.edu