Forum: Software Forum

Post: Requesting customer feedback about user and password management

Mon, 19 May 2008 13:36:23 EDT

We are trying to streamline the management of user accounts and user passwords and have some questions about current usage patterns:

What method are you using to add user accounts for the SiCortex cluster nodes? Are you using passwd/shadow file manipulation, LDAP accounts added to SSP ldap database, a central site LDAP proxied through SSP, or some other method. Please describe.

What method are you using for changing SiCortex cluster node user passwords?

How would your prefer to do user account management and password changes on the SiCortex system?

View this individual post's RSS Feed

Post: Commercial FE Software

Mon, 12 May 2008 17:24:29 EDT

Is there any feedback on running commercial finite element software such as ANSYS, ABAQUS, LS-DYNA etc. with explicit solutions on your machines? any benchmarks?

Thanks,

Ahmet

View this individual post's RSS Feed

Post: more ways to run MPI

Mon, 21 Apr 2008 11:23:21 EDT

Thanks for the previous help with MPI. I'm now trying another experiment where I want to run my program as a worker on the SiCortex machine and have the master reside on my Windows box. Since SLURM
is responsible for managing the MPI processes, how would I go about doing something like that? On a Windows setup, it would be something like this:

mpiexec -n 1 \\teds\d_drive\mywork\cmochk\cmochk_alpha.exe -zcheck -mpi ... -path \\juliet\s_drive\cmo_cdi -cdu_path

: -n 4 -host blade2 -map s:\\juliet\s_drive \\teds\d_drive\mywork\cmochk\cmochk_alpha.exe -zcheck ...

View this individual post's RSS Feed

Post: SVN Emerge failed

Wed, 16 Apr 2008 12:12:50 EDT

We're trying to emerge subversion on an SC072-PDS, but we get the following build error:
/usr/bin/scg++: line 8: -shared: command not found

Is this a known problem? Is there a way to fix it?

Jim

View this individual post's RSS Feed

Post: Panasas file system

Fri, 11 Apr 2008 13:23:54 EDT

Can the Panasas file system product work with the SiCortex system SC5832 ? Are their any special integration issues to be aware of.

Typically, one installs the Panasas DirectFlow client on the compute nodes and away you go.

What about support for pNFS ?

Oh - and BTW is the SC5832's 17Kw for 5.8 TFLOPS for real ?

Thanks...

View this individual post's RSS Feed

Post: running mpi

Thu, 10 Apr 2008 15:35:22 EDT

I'm not clear how I should run my MPI job with srun. On my 4 processor windows box, I would do something like this:

mpiexec -n 4 program.exe ...

What is the comparable expression if I want to run 4 processors on 1 node?

Thanks,

Ted

View this individual post's RSS Feed

Post: Pathscale + OpenMP + Large Memory

Wed, 09 Apr 2008 14:55:48 EDT

When the pathscale compiler has OpenMP enabled, is memory allocated differently?

The below program, when compiled with:
> pathcc -Wall test.c
works correctly. When compiled with:
> pathcc -mp -Wall test.c
the 3rd malloc() call fails.

I'm allocating 3GB of memory, which fits fine when OpenMP isn't enabled.

#include
#include
#include

int main(void) {
float *b0, *b1, *b2;
int max = 1<<28;
int count;
b0 = (float*)malloc(max*sizeof(float));
b1 = (float*)malloc(max*sizeof(float));
b2 = (float*)malloc(max*sizeof(float));
if ( !b0 || !b1 || !b2 ) {
fprintf(stderr, "Unable to allocate %ld (bytes). Bad\n", 3*max*sizeof(float));
fprintf(stderr, "pointers: %p\t%p\t%p\n", b0, b1, b2);
exit(1);
}

for ( count = 0 ; count < max ; count++ ) {
b0[count] = b1[count] * b2[count];
}

free(b0);
free(b1);
free(b2);

return 0;
}

View this individual post's RSS Feed

Post: Performance Counters

Wed, 09 Apr 2008 08:35:01 EDT

I see from the Programmer's Guide that there are a significant number of Performance Monitoring Counters available on the SiCortex system. Are there instructions somewhere for how to directly access these? At the moment, I'm most interested in accessing CPU_CYCLES, as some of our internal benchmarks rely on the 'rdtsc' instruction of x86 systems to get a count of the number of clock cycles certain operations take.

View this individual post's RSS Feed

Post: MPI, Threads & Signals

Wed, 09 Apr 2008 07:57:49 EDT

I noticed in reading the Programming Guide, that the SiCortex MPI implimentation only supports MPI_THREAD_FUNNELED. I have an application that runs multiple threads sharing a large read-only memory space and needs to communicate between nodes. My question is how do I construct a process that can simultaneously wait for an MPI event and a thread finishing. If I sit in an MPI_Wait() call, I'll never see a thread finishing, and if I wait in a pthread_cond_wait(), I'll never see MPI events. If one thread could send an MPI message to another thread, I could use that to wake the process, but that doesn't seem possible with this implimentation.
Is there some way to use signals to wake a process that's in an MPI_Wait() call? The MPI reference doesn't define the interaction between signals and MPI.

View this individual post's RSS Feed