Subject: Re: Multiprocessor with NetBSD ?
To: Johnny Billquist <firstname.lastname@example.org>
From: Robert Elz <kre@munnari.OZ.AU>
Date: 06/04/2001 16:48:04
Date: Mon, 4 Jun 2001 11:16:37 +0200 (CEST)
From: Johnny Billquist <email@example.com>
| As for the granularity of locking, that don't have anything to do with SMP
| or not, only performance.
Yes, exactly - those claiming that SMP needs fine grained locking
in order to be called SMP are simply wrong ... all that is needed
is for all of the processors to be treated equally (as much as is
possible). After that you're just into questions of quality of
implementation, for which there is rarely ever any one metric by
which one implementation can be said to be unilaterally better than
Consider an analogy with multiprocessing on a single processor.
That's generally defined as being when a process has the illusion
of running on the processor alone, while other processes running
concurrently have the same illusion.
Now, that's generally done via one means or other of implementing a
time quantum - every so often the processor is switched to another
process, so each gets some fraction of the available CPU time.
I doubt that anyone would argue that the time quantum has to be
infinitely small for this to still be called multi processing,
in fact, making it too small hurts performance (all the processor ever
does is context switch, no applications do any work at all).
The best value to use depends upon circumstances, and while making
it too large is generally not good, there's no magic point at which
the system stops being a multi processing system.