[Beowulf] Stroustrup regarding multicore

Lux, James P james.p.lux at jpl.nasa.gov
Wed Aug 27 09:42:04 PDT 2008

-----Original Message-----
From: Steve Herborn [mailto:herborn at usna.edu]
Sent: Wednesday, August 27, 2008 9:33 AM
To: 'Vincent Diepeveen'; 'Tim Cutts'
Cc: Beowulf at beowulf.org; Lux, James P
Subject: RE: [Beowulf] Stroustrup regarding multicore

Perhaps his judgment isn't "universally" objective (not influenced by
personal feelings, interpretations, or prejudice; based on facts), but could
very well could be objective (and his version of reality) in his little part
of the world.  Being at a University I've run across more of what he said
then your particular comments on the subject.

However, that being said I would think that it is usually easier to teach a
Scientist to code, then a coder the PhD level of the science.


I would say that you shouldn't be teaching the Scientist to code. You should be teaching them to write good requirements documents & test cases that someone who is skilled at coding can use.  That way, you can leverage the years of tuning experience and down and dirty knowledge of the computational platform.  Giving the Scientist background information so that they can effectively ask for things is another story (i.e. make them an intelligent consumer).

And, even if the Scientist does wind up coding, they'll produce better code if they didn't just do it in a ad-hoc "let's go in and modify what 30 years of grad students have done and see what happens".. Not that you're not going to wind up modifying the 30 years of heritage, but at least, you'll have a clearer picture of what, exactly, you're trying to accomplish and how you'll know that you got there. That's always nice when you have to justify how you spent all that money.

{This comment is given, of course, wearing my software development manager hat.  I reserve the right to hack up my own stuff, at my own risk, of course.}

Jim Lux

More information about the Beowulf mailing list