Need advice on parallelizing fortran 90 code
in [Fortran]
|
Prev: Type Casting INTEGER*4, DIMENSION 2 into 1 REAL*8
Next: Q Dislin: 3d plots with axis crossed in a point
From: octaedro on 13 Apr 2008 14:37 Hi folks, I am a PhD student in economics that uses quite computationally intensive models (under economists standards). I am working on an Intel Core 2 Duo T7200 @ 2GhZ and 2GB RAM. I am also using Compaq Visual Fortran v6.6. The nature of my research involves dynamic programing over large state spaces and calibrating such models. As usually I deal with non smooth problems I sometimes deal with waiting days or even weeks to get results and check if I made some mistake. I was wondering if anyone could offer some advice on whether I can use what I have to write parallel code and how does it work (or any good reference). I know that I am being a little bit fuzzy but I would appretiate if you could give me any advise. Thank you Jorge Alonso
From: Jan Gerrit Kootstra on 13 Apr 2008 14:46 octaedro wrote: > Hi folks, > I am a PhD student in economics that uses quite computationally > intensive models (under economists standards). I am working on an > Intel Core 2 Duo T7200 @ 2GhZ and 2GB RAM. I am also using Compaq > Visual Fortran v6.6. > The nature of my research involves dynamic programing over large state > spaces and calibrating such models. As usually I deal with non smooth > problems I sometimes deal with waiting days or even weeks to get > results and check if I made some mistake. > I was wondering if anyone could offer some advice on whether I can use > what I have to write parallel code and how does it work (or any good > reference). I know that I am being a little bit fuzzy but I would > appretiate if you could give me any advise. > Thank you > Jorge Alonso Jorge Alonso, If you do not give us an example of the code or what type of algorithems you use and advise is hard to give. Normally algorithems with a lot of do loops are easy to paralellize by OMP capable compilers. You will need to upgrade to an Intel Visual Fortran compiler or another parallelizable compiler. Kind regards, Jan Gerrit Kootstra
From: Gib Bogle on 13 Apr 2008 18:00 octaedro wrote: > Hi folks, > I am a PhD student in economics that uses quite computationally > intensive models (under economists standards). I am working on an > Intel Core 2 Duo T7200 @ 2GhZ and 2GB RAM. I am also using Compaq > Visual Fortran v6.6. > The nature of my research involves dynamic programing over large state > spaces and calibrating such models. As usually I deal with non smooth > problems I sometimes deal with waiting days or even weeks to get > results and check if I made some mistake. > I was wondering if anyone could offer some advice on whether I can use > what I have to write parallel code and how does it work (or any good > reference). I know that I am being a little bit fuzzy but I would > appretiate if you could give me any advise. > Thank you > Jorge Alonso I would say that DP is a good candidate for parallelizing, and you could do it on your machine with Intel Visual Fortran and OpenMP, which is a shared memory parallel computing system. But the effort involved in getting such code to run on a dual core machine, with a best-case speedup of 2x, hardly seems worthwhile to me. For jobs the size of what you describe, in my opinion you'd be better advised to used distributed parallel programming, e.g. with MPI, which can utilize multiple computers, but this is a major step in programming complexity.
From: glen herrmannsfeldt on 13 Apr 2008 18:36
octaedro wrote: > The nature of my research involves dynamic programing over large state > spaces and calibrating such models. As usually I deal with non smooth > problems I sometimes deal with waiting days or even weeks to get > results and check if I made some mistake. > I was wondering if anyone could offer some advice on whether I can use > what I have to write parallel code and how does it work (or any good > reference). I know that I am being a little bit fuzzy but I would > appretiate if you could give me any advise. Many dynamic programming problems can be partitioned such that no communication is needed between the different parts. Others can be run in batches with the output from one batch fed as input to a later batch. Otherwise, you will need to say more about the problem you are working on. -- glen |