PARDISO Parallel problem

[ray_fl]

Hey all,

I am using PARDISO in IMKL 8.01 for my program on Pentium Dual Core 3.2 G (no HT) platform.

The program is win32 console by VC.net 2005.

I followed the instructions by setting OMP_NUM_THREADS = 2.

I tried my program using IA32 lib on winxp pro and EMT64 lib on Winserver 2003 x64 and got the same strange results.

From the solving statistics below, most of the time spent on step 1: reorder However, the cpu usage of this step is only 50%.
Step 3 can use 100% cpu. I am not quite sure about step 2 cpu usage.

So from here seems little meanings using multiprocessor for PARDISO.

My questions are:

1) How can I make full use of dual core resource for sparse matrix solving in IMKL?

2) Could you explain my results?
Did I neglect some important settings supporting multi-threads
or PARDISO reorder step can not be parallelled?

3) There are some link warning when I use VC.net 2005 as below:
Does it relate to my results. I don't have such warnings in VC 6.0.

Any suggestion to speed the program is welcome!

Thanks.

Regards,

Ray

***************
Link warnings
***************

1>libguide.lib(kmp_csupport.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_sched.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_ftn_cdecl.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_dispatch.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_runtime.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_global.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_lock.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_debug.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(z_Windows_NT-586_util.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_alloc.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_i8_dispatch.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_i8_sched.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_ssp.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_statgen.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_error.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_threadprivate.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_utility.obj) : warning LNK4224: /COMMENT is no longer supported; ignored
1>libguide.lib(kmp_ftn_extra.obj) : warning LNK4224: /COMMENT is no longer supported; ignored

**************************************
The solving Statistics
**************************************

================ PARDISO: solving a symmetric indef. system ================

Summary PARDISO: ( reorder to reorder )
================
Times:
======
Time fulladj: 0.251597 s
Time reorder: 1829.120878 s
Time symbfct: 3.429548 s
Time parlist: 0.303471 s
Time malloc : -0.957534 s
Time total : 1837.942776 s total - sum: 5.794816 s
Statistics:
===========
< Parallel Direct Factorization with #processors: > 2
< Hybrid Solver PARDISO with CGS/CG Iteration >
< Linear system Ax = b>
#equations: 3838560
#non-zeros in A: 10111477
non-zeros in A (%): 0.000069
#right-hand sides: 1
< Factors L and U >
#columns for each panel: 80
#independent subgraphs: 0
< Preprocessing with state of the art partitioning metis>
#supernodes: 3194983
size of largest supernode: 1861
number of nonzeros in L 104270391
number of nonzeros in U 1
number of nonzeros in L+U 104270392
================ PARDISO: solving a symmetric indef. system ================

Summary PARDISO: ( factorize to factorize )
================
Times:
======
Time A to LU: 0.000000 s
Time numfct : 106.131564 s
Time malloc : 0.898005 s
Time total : 107.030807 s total - sum: 0.001238 s
Statistics:
===========
< Parallel Direct Factorization with #processors: > 2
< Hybrid Solver PARDISO with CGS/CG Iteration >
< Linear system Ax = b>
#equations: 3838560
#non-zeros in A: 10111477
non-zeros in A (%): 0.000069
#right-hand sides: 1
< Factors L and U >
#columns for each panel: 80
#independent subgraphs: 0
< Preprocessing with state of the art partitioning metis>
#supernodes: 3194983
size of largest supernode: 1861
number of nonzeros in L 104270391
number of nonzeros in U 1
number of nonzeros in L+U 104270392
gflop for the numerical factorization: 32.926163
gflop/s for the numerical factorization: 0.310239
================ PARDISO: solving a symmetric indef. system ================

Summary PARDISO: ( solve to solve )
================
Times:
======
Time cgs : 5.478609 s cgx iterations 2
Time malloc : 0.055662 s
Time total : 5.727567 s total - sum: 0.193296 s
Statistics:
===========
< Parallel Direct Factorization with #processors: > 2
< Hybrid Solver PARDISO with CGS/CG Iteration >
< Linear system Ax = b>
#equations: 3838560
#non-zeros in A: 10111477
non-zeros in A (%): 0.000069
#right-hand sides: 1
< Factors L and U >
#columns for each panel: 80
#independent subgraphs: 0
< Preprocessing with state of the art partitioning metis>
#supernodes: 3194983
size of largest supernode: 1861
number of nonzeros in L 104270391
number of nonzeros in U 1
number of nonzeros in L+U 104270392
gflop for the numerical factorization: 32.926163
gflop/s for the numerical factorization: 0.310239