eugen at leitl.org
Fri Jan 21 06:57:06 PST 2005
On Fri, Jan 21, 2005 at 09:19:14AM -0500, Mark Hahn wrote:
> otoh, many people reagard "real" parallelism to be much more tightly coupled
> than that. for instance, suppose you're doing a gravity simulation where
> each star in your virtual cosmos influences the motion of each other star.
> MPI is what you want, though you can also do it using shared memory (OpenMP).
> the point though is that you absolutely must think in terms of message
> passing no matter how your parallelism is implemented, because you have so
> much communication.
On a mildly lunatic note, message-passing fits the constraints of the
computational physics of this universe very nicely. When I ask memory
for a word, I send the address message, and receive a contents message.
That's a minimal overhead in terms of signals propagating and gates
switching required, and whether you see it, or not, depends on your
requirement profile. SCI internode can be in principle very
close to accessing physical memory. The latency would be largely
relativistic, and depend on the distance.
If I have a physical system, there's no way how a part of it can influence
simultaneously all others (in a meanigful way, that we can tell apart from
random way) in a relativistic universe. Depending on your temporal step size,
and the size of the system broadcast or a locally-coupled communication
pattern might be appropriate.
> message passing is not an overhead, but rather a consequence of what data
> your problem needs to exchange. if you have a lot of data exchange, and
> do not think in terms of discrete packets of data collected and sent where
> needed, your performance will SUCK.
Eugen* Leitl <a href="http://leitl.org">leitl</a>
ICBM: 48.07078, 11.61144 http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 198 bytes
Desc: not available
More information about the Beowulf