How hard could it be?

It happens at every grid or distributed computing conference... a new physicist (one usually who already knows something about computing) shows up and makes a remark to the effect, how hard could this be?

As physicists, we are trained (& indoctrinated) that we can solve anything.  That is how we learn physics, by solving problems.  I have preserved my carefully written solutions to  Messiah's Quantum Mechanics, and probably more valuable, Jackson's Electrodynamics, the bible of electricity and magnetism for most every physics graduate student. (Once during a colloquium I had J.D. Jackson autograph my copy of "Jackson", people thought it was groupie-ish, but for me, I really loved and enjoyed that book... I also had an excellent prof, P. Shanley).

The same holds true for computing.  It so holds true.  We usually have to develop our own stuff, and in most cases things eventually get built and work out just fine.  And these are considerable achievements in some cases.   However, we've got no market on building large scale things for computing in science, and actually a lot of (gulp) computer scientists who know what they're doing when it comes to engineering production scale software (in addition to innovating new ideas).  They usually warn us to  "not try this at  home".  Problem is, we have no choice,  we must try this at home.  By the way, how hard could it be?

In practice, I think there is a balance to be struck here and blanket generalizations are pretty dangerous.

These are some things I was thinking about listening to the Condor guys talk about some of their new Condor-C software.  All kinds of cool things for job routing, delegating credentials, gliding in remote schedulers, really neat stuff that is actually working today.

Here is Todd Tannenbaum showing some of the arcane details of the new software - I think he was talking about speeding up a matchmaking routine.. interspersed with mini-discourses on "pre-emption rank", "semantic gaurantees" and "first class leases" and "repudiation" etc.  Turns out our grid (OSG) at the moment doesn't support first class leases (natively), but we'll be able to effectively do it by adding a layer of indirection (CS people know that every software problem can be solved that way!).

(Guys: here are snaps of the whiteboard images, for reference.)

At any rate, the workshop today was great - in addition to the Condor-C "strike team", there was a small gang of CS graduate students from U of Iowa (working for Shaowen Wang), physicists & computer scientists from CDF and CMS from Fermilab, all figuring out how the technology is going to work for them.  Thanks to Dan Bradley  and Nate Mueller (U Wisconsin physics, CS) for organizing this very short but productive workshop.

Gotta hit the road..car wash scene from Madison.

Comments

[A.P.]

From the instant I saw the "Octopus Car Wash" photo, I knew this must have taken place in Madison. Go Badgers!