multisplit in networks

General issues of interest both for network and
individual cell parallelization.

Moderator: hines

Post Reply
ammu
Posts: 9
Joined: Tue Feb 14, 2012 12:10 am

multisplit in networks

Post by ammu »

hi,
i followed the 'Fully Implicit Parallel Simulation of Single Neurons (Hines et al. 2008 ' model and applied the concept in several other single neuron models. And using the same concept i scaled it up to my network of 45 cells. Interestingly the method works when i use np=1 [mpirun -np 1 <nrniv path> -mpi -NSTACK 2000 <filename>]. However the method fails to work when issued the same command with np=4 [mpirun -np 4 <nrniv path> -mpi -NSTACK 2000 <filename>]. The following is the error which resulted when run with np=4

0 /mirror/neuron/nrn-7.3/x86_64/bin/nrniv: Segmentation violation
1 /mirror/neuron/nrn-7.3/x86_64/bin/nrniv: Segmentation violation
1 in init.hoc near line 91
1 model()
^
1 finitialize(-70)
0 in init.hoc near line 91
1 init(0 model()
)
1 stdinit( )
1 statrun( 100 )
and others
application called MPI_Abort(MPI_COMM_WORLD, -1) - process 1

Could you let me know where the problem could be? And also i wanted to know what does NSTACK refer to ?
hines
Site Admin
Posts: 1682
Joined: Wed May 18, 2005 3:32 pm

Re: multisplit in networks

Post by hines »

NEURON's Hoc interpreter has a fixed maximum size stack specified early after launch. It's default size is 1000 in src/oc/code.c
#define NSTACK 1000 /* default size */
The -NSTACK <val> changes this. It is not generally necessary to use it unless one gets a stack error that is determined not to be user bug because you really need a large stack because of a large depth of recursive function calls.

With regard to the multisplit problem with -np 4, can you send me all the code (hoc, ses, mod files in a zip file) needed to reproduce the problem and I'll look into it. Send the zip file to michael dot hines at yale dot edu .
Post Reply