VERY Large Matrix Computation
Author Message
VERY Large Matrix Computation

At the Research Institute I work at, we are running fortran
programs that solve at least 500,000 unknowns sometimes a million.
As you can imagine, it takes a long time for the computation to
finish. What I as wondering is:

I guess part of this long computation time can be blamed on the
swapping process. I was wondering if there isn't any special methods(math)
or implemeations (like linked lists) known which will speed things up.
The Fortran-master I work with uses some sort of Gausian method I guess.
I know that for a Gausian calculation to happen, all the matrix
has to be in memory which is I think where the problem occurs. Data
is read from file, because everythins can't be placed to memory it's swapped back to
hard-disk. Then when it's needed, it's retrieved back again. So this
makes 3 disk accesses. Can't this process be optimized by reading
chunks of the matrix into the memory as it is needed?? (probably
by using another method)

*****************************************************************
*                                                               *
*       Burak Bayramli                                          *
*                                                               *
*       Stevens Institute of Technology                         *
*                                                               *
*****************************************************************

Sat, 23 Aug 1997 23:03:55 GMT
VERY Large Matrix Computation
We have commercial high performance Out-Of-Core Matrix solvers for various

Computation>Date: 7 Mar 1995 15:03:55 GMT

Quote:
>At the Research Institute I work at, we are running Fortran
>programs that solve at least 500,000 unknowns sometimes a million.
>As you can imagine, it takes a long time for the computation to
>finish. What I as wondering is:
>I guess part of this long computation time can be blamed on the
>swapping process. I was wondering if there isn't any special methods(math)
>or implemeations (like linked lists) known which will speed things up.
>The Fortran-master I work with uses some sort of Gausian method I guess.
>I know that for a Gausian calculation to happen, all the matrix
>has to be in memory which is I think where the problem occurs. Data
>is read from file, because everythins can't be placed to memory it's swapped
>back to
>hard-disk. Then when it's needed, it's retrieved back again. So this
>makes 3 disk accesses. Can't this process be optimized by reading
>chunks of the matrix into the memory as it is needed?? (probably
>by using another method)
>*****************************************************************
>*                                                               *
>*       Burak Bayramli                                          *
>*                                                               *
>*       Stevens Institute of Technology                         *
>*                                                               *
>*****************************************************************

Sat, 30 Aug 1997 21:24:58 GMT

 Page 1 of 1 [ 2 post ]

Relevant Pages