[Beowulf] computing on Altix?

[Andrew Piskorski]

On Fri, Sep 09, 2005 at 08:12:37AM +0100, Hongmei He wrote:

> I submit a task to Altix 350 share memory parallel machine.  The
> task is to compute a graph, and rquests 6, 9, 12, 16 processors
> respectively. The results show that the running time is decreasing
> with the rise of the size of processors for same graph, while each
> processor runs same program, and communicates with fixed number of
> neighbours periodically. why?

Google for "superlinear speedup".  Most likely, as you split up your
fixed problem size among more processors, more and more of it fits
into the processor cache, where it runs much faster due to fewer main
memory accesses.

The more interesting question is how many cpus can you use your
existing code on and still get worthwhile speedup?  Since you are on
an expensive ccNUMA machine, your communication overheads are probably
low, and the answer might be, "all the cpus this machine has".
Consider yourself lucky...

A similar interesting question is, would you stil get similar speedup
on a cheap gigabit ethernet Beowulf cluster?

-- 
Andrew Piskorski <atp at piskorski.com>
http://www.piskorski.com/