some newbie queries about parallel code

[jim_cox]

Hello world

We areinvestigating upgradingto the Intel Fortran compiler

At present we have a significant code base inFortran(Micro$oft v4) with some C++ routines linked in

We are hoping to be able to gainsome speed with parallelisation of code

So some questions if I may...

Is this a realistic expectation?

If so what sort of gains can be expected?

Just how easily can code be converted?

What disadvantages can we expect to see?

At present our code runs on win2K, XP and Vista - from the compiler documentation I am given to understand that it may not work on win2K - is this correct?

We would be grateful for any advice, hints, tips and war-stories

advTHANKXance

Jim Cox

=mjc=

.

[TimP]

The first advice is to read relevant posts on this and other forum sites.
Fortran code with 2 or more levels of DO loops often is easily parallelizable, by OpenMP or, in the simplest cases, -Qparallel. Intel Thread Checker is a convenient tool for diagnosing errors in OpenMP parallelization.
In order to parallelize on a cluster of computers, MPI code usually is required, and often hardware communications facilities such as InfiniBand. MPI tools for Windows have lagged those available for linux.
Applications suitable for OpenMP frequently speed up by more than a factor of 3, on current dual quad core platforms. This speedup will improve in the near future, but emphasis is likely to be on 64-bit OS.
Win2K may be superior to 32-bit XP in its support for parallelization on multiple cores, although its support for HyperThreading is rather poor. Certain Intel tools, such as VTune, no longer support Win2K, nor does Microsoft support it to the extent required for hardware or software vendors to support it. Any 32-bit OS limits the opportunity for parallelization of large applications.