[Beowulf] Stroustrup regarding multicore

Jonathan Dursi ljdursi at gmail.com
Thu Aug 28 09:40:32 PDT 2008


2008/8/28 Peter St. John <peter.st.john at gmail.com>:
> I think a physicist programming is like an astronomer grinding lenses (maybe
> nobody does that anymore).

There is an even more direct analogy which I find useful -- experimental
physicists.  (Or chemists, or. what have you).  To do cutting-edge
stuff you need to build some of your own stuff -- if what you wanted to
do could be done entirely with off-the-shelf equipment,it'd probably
already be done.   So you have to build your own equipment, meaning
that there has to be people who know the science and people who know
equipment-building techniques.

In a huge collaboration (like big particle physics experiments) you
really do have the money to have people ranging from  pure builders and
pure theoretical scientists with relatively small overlap of knowledge.
But in the vast majority of cases, that doesn't happen, and you have to
build teams of  people who each know some degree of both.

It's important not to lose track of the fact that it is (or should be)
the science driving these `numerical labs'.   So yes, when the dollars
are scarce it is more important to have people who know the science than
know how to code well; at least then then you can actually solve the
science problem.  But then you end up solving it inefficiently or in
a way that makes it needlessly difficult to generalize  to solving the
*next* science problem.  Much more progress, and faster, is made if the
researchers know how to do both well, or at least have people to turn
to whose knowledge complements theirs.

Experimenters have had a good century or two head start to figure out
how to train experimental scientists; it shouldn't surprise us that
we're still figuring out what the right combination of people and skills is to
staff computational science `labs'.   But we'll get there.   (And hopefully
sooner than later.  If experimental science were done with the same approach
that computational science is done in my field, we would still think that
light sailed through an aether...)

  Jonathan

--
Jonathan Dursi
ljdursi at gmail.com



More information about the Beowulf mailing list