10/2 Is there any advantage multitasking-wise of a system with N processors
over a system with one processor that runs N times as fast? (Of
course there are other factors like N-slow-processor systems can be
available before the N-times-fast processor itself is, and maybe N
slow chips are cheaper than one fast chip.) Thanks.
\_ MP's are worthless unless you have an OS that supports SMP or
something along those lines. Solaris supports it, Linux does
too but I don't know how well, FreeBSD will pretty soon, and
NT does but we all know how reliable that is. If you're using
Win95/98 don't bother with more than one proc. -jeff
\_ The more you scale a multiprocessor system the more bus traffic
you'll encounter so it will only scale up to a certain point.
This can be fixed with a larger cache on each processor but this
simply results in more cache incoherency causing even more bus
traffic.
\_ The fast processor can simulate the N chips in just about all
situations. Whereas N chips cannot simulate the single fast
chip in all situations.
On the other hand, if you have really really good I/O subsystems
if due to more
designed for multiprocessor,
and lots of multitasking, the N processor system can be a win.
Assuming they have good cache, etc.
\_ you must be this tall...
\_ this question depends on your workload. yes, there can be an
advantage due to saved context and keeping your TLBs. if the work
can't be parallelized you may end up losing on an MP. none of
this really matters unless you are doing serious crunching .-aaron
\_ it depends on why the single processor is N-times faster.
If due to more execution units, N procs gives you more regs,
if due to more registers, N procs gives you more exec units,
if due to higher cache bandwidth, etc. etc. There will
be situations where the N processors have the upper hand. -nick |
http://csua.com/feed/