topic Hi, in Intel® oneAPI Math Kernel Library

How to config the PARDISO OOC file and problems with SORTQQ

Gators_vs__Sundevil — Wed, 15 Apr 2015 20:12:39 GMT

Hi,

I have a question on how to config the OOC file for PARDISO in order to use the OOC mode.

I'm trying to solve a very huge system, may take up to 500G RAM. I have a Xeon workstation with 250G RAM and 4T hard disk running Ubuntu OS. It is obvious that I should use the OOC mode in order to get the equation system solved. But for the OOC, as far as I know, there are some parameters need to be specified. And the following are the values I have used: MKL_PARDISO_OOC_MAX_CORE_SIZE=200000, MKL_PARDISO_OOC_MAX_SWAP_SIZE=0, MKL_PARDISO_OOC_KEEP_FILE=0. There were a segmentation fault error after some temporary files stored in the disk. For the same equation size, this problem kept occur every time I tried to run the code, but the sizes for the temporary stored files were different.

So, what may be the reason for the segmentation error? If this is due to the OOC parameters setting, how should I configure the file? Or, there are some problems with the harddisk or any other OS setting?

Another question is about the SORTQQ function. The input parameters I'm using is SORTQQ(array, size, SRT$INTEGER4). The entries of the array are integers. This function with the input parameters worked fine on a windows OS. But there was a warning message when on a Ubuntu OS: "the data type of the actual argument does not match the definition". I tired to change the data type to SRT$INTEGER8 and the system gave me the same warning. How should I solve this problem?

Any suggestions on the above problems are much appreciated.

Hi,

Alexander_K_Intel2 — Thu, 16 Apr 2015 02:34:17 GMT

Hi,

Pardiso question - can you set msglvl to 1 before first pardiso call and all test that pardiso print on the screen to me? It could help to find rootcause of problem.

About SORTQQ - could you sent to me small reproducer of this problem? Look like there is an mix of type of parameter that correspond to different local variables but it hard to guess without code.

Thanks,

Alex

Quote:Alexander Kalinkin

Gators_vs__Sundevil — Fri, 17 Apr 2015 20:09:51 GMT

Alexander Kalinkin (Intel) wrote:

Hi,

Pardiso question - can you set msglvl to 1 before first pardiso call and all test that pardiso print on the screen to me? It could help to find rootcause of problem.

About SORTQQ - could you sent to me small reproducer of this problem? Look like there is an mix of type of parameter that correspond to different local variables but it hard to guess without code.

Thanks,

Alex

Hi Alex,

Thanks for your reply.

AS from the printed statistical information, the A matrix (AX = b) is zero or negative pivot. I need to check the A matrix.

For the SORTQQ, sorting any integer array using the SRT$INTEGER4 or 8 parameter will always give me the same warning message on Ubuntu. But the warning only appears during the compilation, the function works fine since the solution after ignoring the warning is correct. The warning is really annoying.

Thanks.