>In article <Pine.LNX.3.95.970920163727.12141A-100000@lifesaver>, Andreas
>Kostyrka <andreas@ag.or.at> wrote:
>>Win/NT SMP sucks. C't (a german magazine tested SMP boxes) found, that it
>>doesn't scale very much, because it seems that interrupts are done only on
>>one processor. Linux 2.0/SMP sucks probably even more :(, as it has a
>>global spin lock :(((((
>InternetWeek did an article on different OS's acting as web servers a
>couple of weeks ago. NT actually got slower with a second CPU.
This is not a surprise. It's well known that machines which are IO
bound can often suffer greatly from having too many CPUs. When I was at
university one of the UNIX servers was taken from 8 CPUs to 2 to improve
it's network and file performance. The machine was only allowed to have 2
CPUs because some people were doing parallel processing research on it,
otherwise it would have gone down to 1 CPU.
If your OS is spending all it's time processing interrupts/DMA etc then
having multiple CPUs fighting over this is unlikely to improve performance,
even if interrupts can be processed on multiple CPUs. If only 1 CPU can
handle interrupts and it has to ackquire global locks to do some of it's
work then things may be really bad.
Does anyone have any reports of a Linux system with 2 CPUs running
faster than a system with a single CPU?
-- ----------------------------------------------------------- In return for "mailbag contention" errors from buggy Exchange servers I'll set my mail server to refuse mail from your domain. The same response applies when a message to a postmaster account bounces. "Russell Coker - mailing lists account" <bofh@snoopy.virtual.net.au> -----------------------------------------------------------