On Fri, 7 Apr 2000, Ian C.Sison wrote:
..
> There was an article i read once that mentioned the fact that
> distributing threaded code over several processors actually degrades
> performance rather than improves it. Reason? Context switching
> between threads and processors breaks the caches of the CPUs enough to
> provide a noticable performance hit. So much so, that you'ld be
> better off with a uniprocessor box.
Thats true for pathological cases (IBM pointed this out). But for general
threaded apps particularly if you have a large-enough cache, the
combination of the multiprocessors and cache affinity prevents thrashing.
> > The context-switch latency on Linux is much less than on other
> > OS's (see McVoy's lmbench) so there is no time savings in avoiding the
> > fork, unlike say Solaris where a thread switch takes 1/100th the time of a
> > full context switch.
>
> I'm not sure about this one, and neither do i have benchmark results
> proving that the assertion i read was really true. \8)
---------------------------------------------------------------------
Orlando Andico <[EMAIL PROTECTED]> +63 (2) 937-2293
Mosaic Communications, Inc. +63 (917) 531-5893
Any sufficiently perverted technology is indistinguishable from Perl.
-
Philippine Linux Users Group. Web site and archives at http://plug.linux.org.ph
To leave: send "unsubscribe" in the body to [EMAIL PROTECTED]
