tlug: Re: linux cluster

[Darren Cook <darren@example.com>]

>[If] your main bottleneck is CPU <-> memory bandwidth. This
>implies two things:
>
>* Buy the biggest cache you can get

Pentium-II seems limited to 512Kb cache. The next Intel chip is supposed to
have larger cache (and be better designed for parallel processing), but it
is going to be *much* more expensive. 

>(and write your
>  application cache-friendly)

What does this mean? If it's answered in the great link (
http://www.mcs.anl.gov/dbpp/ ) you gave yesterday, ignore this question, as
I'm working through that site. :-)

>* don't use two-processor boards.
>
>The two processor will only end up competing for bus access,
>instead of doing your float calculations.

I also read somewhere that even if the data set is small enough to fit in
the processor cache, on a dual processor machine they have to spend time
comparing the two caches to make sure the processors are not altering the
same memory. Single processors in a cluster won't have this problem.

Darren



---------------------------------------------------------------
Next Nomikai: 20 November, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next Meeting: 12 December, 12:30 Tokyo Station Yaesu central gate
---------------------------------------------------------------
Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp