Networks with RxD don't get parallelized properly

[salvadord]

Truccolo:
Things run fine when running on a single processor: 4 diffusion sources, etc. However, when running with 3 processors, we are back to seeing only 2 sources instead of 4.

3 processors:
"Instantiated 0 cells of population I2" and "Instantiated 0 cells of population I4",
1 processor:
"Instantiated 1 cells of population I4" and "Instantiated 1 cells of population I2"...

Craig:
I've got a network of 11,250 cells, I'm running running mpiexec -n 6 python init.py , and it prints Creating network of 1 cell populations on 1 host... and Number of cells on node 0: 112506 times each. So it seems like I'm just running init.py 6x rather than actually parallelizing anything . . . which maybe isn't the same problem

[salvadord]

Discussion:

Craig 4:40 PM
yeah, i had the same problem when getting rid of extracellular diffusion
ramcdougal 4:42 PM
Then something is wrong with how netpyne sets things up because intracellular rxd only happens within a given compute node and doesn't use MPI at all
4:44
If I had to guess... Maybe reactions or regions are only being defined on some of the nodes?
4:45
Where's the netpyne code that creates things? Are we assuming any particular version of NEURON?
4:47
with a recent enough version (7.7 maybe?) everything can be done on a per cell basis... In older versions, you'd have to be careful about what sections are where

Craig 4:49 PM
so I had the issue with the netpyne example: examples/rxd_net, and using NEURON 8.0.dev