Good question. The error message isn't very explicit, so here's how to decode it.
Syntax errors typically occur when the model source code is first being parsed. They often contain a useful clue as to the underlying cause, in the form of the actual statement that caused failure. This
Code: Select all
1 nrniv: syntax error
1 in sj10-cortex.hoc near line 202
1 insert ca
means that host 1 ran into difficulty with an "insert ca" statement while parsing a file called sj10-cortex.hoc.
"insert" statements are used to assign biophysical properties fo sections, and failure of an insert statement generally means that NEURON doesn't know about the mechanism that is supposed to be inserted. This will happen if the mechanism is specified by NMODL code in a mod file that has not been compiled, or if it is specified by a Channel Builder whose ses file has not already been parsed. Since the model runs fine when autolaunched, we know that the problem can't be a missing Channel Builder, because the principal difference between autolaunch and you launching the model from the command line (or double clicking on its mosinit.hoc file) is that the autolaunch process starts by compiling mod files.
The solution is for you to compile the mod files yourself, before trying to launch the model. But how should you do this?
Many models in ModelDB come with a text or html file that describes how to use the code. This particular model's readme.txt starts with somewhat detailed, but very welcome, descriptions of the various model source code files. Just before the end of readme.txt is a short, easily overlooked section with "Usage instructions" that tell how to compile the mod files and launch a simulation
After unzipping the attached file (on the parallel cluster master) and
cd'ing to the created folder compile the mod files with the command
and then type
mpirun -n 4 nrniv -mpi Batch.hoc
replacing the 4 above with a number of processors that you have
That should take care of the problem.
Using 3 cores of my desktop PC, which has a quad core i7 920 running at 2.67 GHz and 8 MB cache, 150 ms of model time take about 20 seconds to complete, so I'd guess that on a machine with similar core speed, 1500 ms of model time would take about 600/n seconds to complete, where n is the number of hosts specified on the command line.