Subject: Re: [netatalk-admins] AS IP speeds
From: Osma Ahvenlampi (oa@spray.fi)
Date: Tue Jan 12 1999 - 04:10:08 EST
Sak Wathanasin <sw@nan.co.uk> writes:
> > I never said that the Mac *hardware* wasn't capable of saturating an
> > ethernet. Clearly a fully multitasking BeOS that has properly written
> > networking capabilities will be able to do much more with the G3
> This doesn't make much sense to me: logically, a pre-emptive multi-tasking
> system, since it has to share the CPU, should be slower than a non-MT system
> which has the CPU 100% dedicated to a particular job (for that job only, of
You underestimate the complexities involved in high-performance
networking. When you include disk-io, network-io, frame construction,
data processing and all the other major tasks that have to be done
here, and remember that "saturating ethernet" means "io-bound", you
should realize that a well-designed multithreaded/multitasking design
almost always significantly outperforms a singletasking design in
IO-bound operation, because the more fine-grained structure allows the
operating system better adjust to run-time enviroment.
And this doesn't even factor in the general bloatedness of MacOS. Even
if it was more efficient to single-task, it certainly isn't more
efficient to have to run many times as much code for the same
operation. And MacOS doesn't dedicate the CPU to a particular job
anyway - in fact, it's cooperative task switch is many times more
expensive than the pre-emptive context switch of a state-of-the-art
operating system.
-- I hope someday a Pope chooses the name Shorty. Osma Ahvenlampi <oa@spray.fi>
This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:16:10 EST