[Users] running Cactus on bigben

Maria Cristina Babiuc maria at einstein.phyast.pitt.edu
Tue May 15 14:15:40 CDT 2007


Hello,

Erik, thank you for your input. 
The arrangement runs fine on my desktop computer, so I'm surprised it could 
run out of memory on a supercomputer. I'm using 4 processors, on two nodes, 
and each compute node has 2 Gbytes of memory.
I tried the option PUGH::storage_verbose = yes, but I need some help in how to 
interpret the output. I am getting a very long list in which the message is:
INFO (PUGH): Switched memory off for variable 'variable_name'  [GV TLs: -11782 
Total Size: 430.99 MB]

Thanks,
Maria


On Tuesday 15 May 2007 14:18, Erik Schnetter wrote:
> On May 15, 2007, at 13:08:59, Maria Cristina Babiuc wrote:
> > Could somebody please help me in running Cactus successfully on the
> > PSC Cray
> > XT3 machine bigben?
> > The wavetoy runs OK, but with my arrangemnt I got the error:
> >  ----- DEBUG: PCB, CONTEXT, STACK TRACE ---------------------
> >
> > SIGNAL #[11][Segmentation fault]  fault_address = 0x0000000000000030
>
> Not being an expert on Big Ben, I nevertheless wager a guess.  A
> segmentation fault can be caused either by an error in your code, or
> by your simulation running out of memory.  Try to estimate how much
> memory you need on every processor, and compare that to the available
> memory on each node.  It may be necessary to use more processors in
> order to reduce the per-processor memory requirement.
>
> If you use PUGH, then the option PUGH::storage_verbose = yes can help
> you.  If you use Carpet, then the option
> CarpetLib::print_memstats_every = 1 may help.
>
> -erik


More information about the Users mailing list