I have a Polywell with two PIII 500Mz cpus with almost 1GB of RAM with the Windows NT4 Worstation OS.

I have been running a Fortran code (NPARC3D...a computational fluid dynamics code) successfully for a few months on this system.

Recently, I have been trying to run two jobs simultaneously from separate directories. The first job runs without any problems but when I try running a second job, I get the following error:

forrtl: severe (22): input record too long, unit 2.

Unit 2 reads the fort.2 file which is the initial flow solution and grid. This is the same fort.2 file that has been read by the first job. I've checked all the input files and found no errors. As a final check, I copied all the files from the first job into another directory and tried to run the second job with those files as input. I still get the same error!

The Fortran code generates temporary files which are stored on the system Temp directory. I'm thought that the two jobs may conflict if they use the same Temp directory so I tried to redirect the temp files from the 2nd job to another location but no luck (I used the enviroment variable TEMP to do this).

Any ideas as to what is happening? I have tried running several jobs in the past on Unix machines with no problems but have little experience with Windows NT.

Hi,

I have been using Windows NT for CFD for quite a long time ( on a DEC alpha, a PC with one processor and a PC with 2 processors) and I never had this problem. I Already had as many as 4 jobs executing in the same machine, without any problem.

Therefore, I would think that the problem you described is not related with NT or the PC, but it has something to do with your code or the compiler.

Regards,

Neyval

Neyval,

Thanks very much for your input.

I suspect that you are right that it has to do with the code but I've run this code on a UNIX machine and had no problems. This code generates temporary files with the status=scratch and puts these into the system TEMP directory. I wondering whether there are any conflicts when two jobs run.

(1). Is this fort.2 file an ASCII file? Can you check the file using different editors to see that it is everywhere correct (Format). (2). Is this fort.2 file got modified as soon as the job started? You can run the first job, copy files over to another subdirectory, stop the first job, delete the first job, and run the second job in this new directory.

John, 1)the fort.2 file is unformatted and cannot be read with an editor 2)the fort.2 file is an input file and should not be changed by the program. I did copy the files from the first job into another subdirectory, stop the first job (not delete it since I can't the purpose of doing so) and ran the job in the second directory; it runs ok, but I want to run two jobs simultaneously from different directories.

(1). Can you run two identical jobs (identical inputs and identical size) at the same time? (2). If not, then it looks like that one job is taking control over the whole machine. (like a DOS program instead of a multi-tasking window program.) (3). If you think the TEMP file is the problem, can you go back to the code and somehow change it to the local subdirectory (or read the file name and path from the keyboard? and recompile the code?

John, 1)I've tried running two identical jobs from separate directories without success 2)I can run one CFD job (from DOS), AND run a grid generator(interactive), AND run Tecplot (which is a post processor), also interactive, simultaneously! 3) I've tried to resolve the temporary files issue by changing the environment variables TEMP and TMP to point to different directories for each job (I use the DOS SET command, in the directory of job launch, to check that the the environment variables have been set correctly. I may have to go through the code and find each occurrence of the OPEN command and make sure that all files are made local to the directory.

(1). I am no expert in NT, but you can go through a simple exercise to see whether a simple Fortran program can be created and run multiple jobs in NT. This will tell you whether the compiler flags are properly set. (2). Write a simple Fortran program with just a long loop to make sure that it will take 30 seconds or so to complete, no I/O. After you have compiled it, run it from two directories to see whether you can run two jobs on the system.

John, I going to run a test with running 2 jobs of NPARC with a one block grid. I suspect the multi-block case generates interface files that may cause the problem.

be sure that you are closing fort.2 soon after you read it. That may make a big impact.

good louck

Tarek,

I'm am not the code author. We got the code from the NPARC Alliance and it's a very long piece of work .

I will try to find where the fort.2 file is opened and read and see if it's closed.

Thanks.

I find it strange that I can run simultaneous jobs on a UNIX machine and not a NT machine.

Does anyone know how NT handles environment variables in relation to a fortran code?

I thing the best thing for you is to check with NPARC developer.

Tareq,

NPARC is no longer supported and the developers have gone on to supposedly a bigger and better code called WIND which seems not to have a fully functional version for PCs

I've found a way around this problem. It's a roundabout solution but it works.

I dowloaded a trial copy of Ataman's TCP Remote Logon software and installed it on the Windows NT machine.

I telnet to this NT machine from another machine (which has Win98) and submit the second job which runs on the second cpu with no problem!

This problem seems to be peculiar to the NPARC/Windows version and may be also a problem with WIND.

(1). So, the files of the second job are written back to this another Win98 machine?

John, the permanent files are located in the same directory where the job is launched. I do not know where the scratch files are located. They are not located in the NT machine's Temp directory where they are usually located!

Another way to run two NPARC WindowsNT jobs simultaneously where both jobs generate scratch files (which are generated when you have a multiblock grid) is to rename one of the NPARC executables! WindowsNT never stops to amaze me with its hidden 'features' .

Another but more elegant technique , now beyond me, is to edit the PIF file of the NPARC executable.

Now I am able to run two simultaneous NPARC jobs where the executable has the same file name! Isn't WindowsNT just wonderful