[TIP] Everybody wants a pony!
C. Titus Brown
ctb at msu.edu
Thu Apr 2 04:38:27 PDT 2009
On Wed, Apr 01, 2009 at 10:57:53AM -0600, m h wrote:
-> Titus, why do scientists care about these platforms? With cheap
-> commodity hardware it seems most people would scale google style.
-> Aren't x86 going to be cheaper/faster? I guess if you are talking
-> about some super optimized fortran compiler, yeah you care, but this
-> is python we are talking about.
Errm, it is every computational scientist's dream to be able to convert
their problem into a map-reduce (or, "pleasantly parallel", or,
"embarrassingly parallel") style problem. Alas, reality has not, so
far, obliged. So some scientists spend a lot of time struggling with
such problems on specialized hardware.
I'm sure the next question will be "like what problems?" so I'll try to
I'm not so up on computational physics, but it's hard to invert a large
matrix on a small computer, no matter how many of them you have.
Genome assembly is tricky to do without a reasonable amount of
preprocessing that looks at the entire data set.
Protein folding is not, generally, pleasantly parallel.
Basically anything that has to look at a non-decomposable data set is
not map-reducable without significant work. There is probably some
theoretical framework that will tell you what kinds of problems fit and
don't fit; I'm not an expert so I'll leave off there.
C. Titus Brown, ctb at msu.edu
More information about the testing-in-python