www.neuron.yale.edu

Hi, Ted,

I have a question about the applicability of parallel simulations with NEURON. From Migliore et al. (J Comput Neurosci 2006), I know setup of a parallel network model uses NetCon as much as possible. Also, each processor integrates the equations for its subnet over an interval equal to the minimum interprocessor synaptic delay. Does it mean parallel simulations cannot be used for a network with graded synapse (no NetCon)? In such a network, the synaptic conductance depends on the presynaptic voltage (or calcium) in a continuous matter. That means all the cells (with connection) have to communicate to each other in each and every time step. If NEURON can still be used to simulate such networks on parallel hardware, will it bring any speed advantage?

Besides, is there a comprehensive beginner level tutorial as to how to setup parallel simulation (with some examples)? Thank you very much!

Best Regards,
GL

First, let me thank you for your interest in using NEURON. I should also mention that your message has been edited slightly for the sake of clarity and to prevent confusion about NEURON's identity. There is no p-NEURON. The name of the program is NEURON, regardless of whether it is being used to execute serial or parallel code. You can name your dog anything you like, but you can't name someone else's dog (you might _call_ it names, lots of them, and some very unflattering, but the dog's name is still whatever its master gave it).

Now for your questions.

NEURON can indeed be used for parallel simulations of network models in which coupling between cells is continuous in time. Parallel simulations may be done with network models that involve any combination of spike triggered synaptic transmission, continuous synaptic transmission, or gap junctions. Let me bring your attention to the ParallelTransfer class, which is documented in the Programmer's Reference. http://www.neuron.yale.edu/neuron/stati ... elTransfer.

That said, implementing network models for parallel simulation is a challenging programming task, and parallelizing networks that have gap junctions is even more challenging because it unavoidably involves many additional technical details. For any network model, however, the first step is to create a serial implementation of the network. This is necessary as a "standard" against which the results of parallelization attempts can be judged.

Thank you very much for your response, Ted. Sorry for the confusion about the name of the program.