Since the release of gridMathematica at the end of 2002, Wolfram Research stated clearly the interest for making parallel computing available for its flagship product Mathematica. Nevertheless, it has not been until the release of Mathematica 7 (six years later) when they have seriously tackled the usability of the parallel computing. Why now?
Curiously enough, the profitability of cloud computing as service on-demand (e.g. Amazon EC2) has only recently convinced IT analysts about how the future technology for business will be, probably seeking for more flexible solutions in a global world with a deep economical instability.
Research cannot be put aside either, and shortages for some scientific and industrial activities will be unavoidable. Wolfram Research has identified this issue and prepare for the forthcoming events, allowing users to benefit from Cloud Computing. Problems like protein folding, DNA sequencing or MonteCarlo simulations are conceivable as “embarrassingly parallel problems” solvable with the help of Amazon EC2. On the other hand, CFDs, heat transfer and other multi-physics simulations could very well enter into the category of “middle-to-highly coupled problems” to be handle by R-Systems supercomputing services. This approach complements the High Performance Computing capabilities of the multicore processors already implemented in the last version of Mathematica. Still gridMathematica remains as a separate product.
I am unaware about how Mathematica plans to merge its per-site License philosophy and the new on-demand accounting, but surely, as stated by Schoeller Porter (technical develop specialist in the Wolfram Partnerships Group) they want to keep their customer satisfied by having it easy, although not necessarily cheap. Moreover the webMathematica interface could very well be the first approach to a complete web-service on demand. Imagine: “Just login, let us compute your problem and pay the bill”. All in one and independent of how difficult or which architecture is underlying.
Probably plenty of users, from medical research institutions, finance stock analysts, scientists simulating biological ecosystems, etc, will be interested. They would rather focus more in their actual domain interest and bother less about how to reach it.
For this solely reason, auditing mechanisms showing how useful the parallelization was or how many resources were used will be mandatory; also open-source solutions showing us what the real balance of price/effort for the whole thing is.