Hi.
I have a couple of questions about Coreneuron before starting to change my current simulation environment.
1) Based on posts on NEURON github, Coreneuron can be run on CPUs and I was wondering if it supports multisplit?
2) If yes, coreneuron + multisplit can provide a boost to the simulation time of single cell models or it will be similar to NEURON + multisplit?
Coreneuron CPU and multisplit
Moderator: hines
-
ted
- Site Admin
- Posts: 6384
- Joined: Wed May 18, 2005 4:50 pm
- Location: Yale University School of Medicine
- Contact:
Re: Coreneuron CPU and multisplit
The primary utility of multisplit is to achieve load balance in network simulations, and that is useful only when the network contains a very few model cells that are very large. Otherwise the performance boost will not be worth the programmer's effort required to restructure your model.
Few single cell models have sufficient complexity for multisplit to do much for run times. One way to find out if your case is an exception, without spending the time and effort of implementing single cell mpi-based multisplit, is to run a thread-parallel simulation on a multicore processor (is anyone using single core processors these days?). No programming effort is required if you use the GUI's ParallelComputeTool (NEURON Main Menu / Tools / Parallel computing), which is probably described somewhere else in the Forum.
If your real aim is to run a large number of simulations of single cell models, you can speed that up with minimal programming effort using NEURON's bulletin-board-style parallelization, and there's also a Python library for that.
Few single cell models have sufficient complexity for multisplit to do much for run times. One way to find out if your case is an exception, without spending the time and effort of implementing single cell mpi-based multisplit, is to run a thread-parallel simulation on a multicore processor (is anyone using single core processors these days?). No programming effort is required if you use the GUI's ParallelComputeTool (NEURON Main Menu / Tools / Parallel computing), which is probably described somewhere else in the Forum.
If your real aim is to run a large number of simulations of single cell models, you can speed that up with minimal programming effort using NEURON's bulletin-board-style parallelization, and there's also a Python library for that.