|Title||Parallel reaction-diffusion simulation in NEURON|
|Publication Type||Conference Paper|
|Year of Publication||2015|
|Authors||Tropper, C., Zhongwei L., Mcdougal R. A., Hines M., & Lytton WW.|
|Conference Name||Society for Neuroscience 2015 (SFN '15)|
|Keywords||SFN, Society for Neuroscience|
We previously developed a stochastic parallel discrete event simulator (Neuron Time Warp), NTW was interfaced with 1D and 3D deterministic reaction diffusion simulators within NEURON. The objective of using stochastic simulation is to portray the variable effects of a small number of molecules in, for example, a dendritic spine. We have developed a threaded simulator, NTW-MT. The reason for developing the threaded version is to take advantage of multi-core architectures employed in parallel machines so that we can increase the speed of a simulation and to be able to simulate larger models. One thread takes care of communications while the others take care of computation. Communication between threads makes use of shared memory if the threads are in the same process and by MPI if they are in different processes. A multi-level queueing stucture is employed for the priority queue. Both NTW and NTW-MT make use of a stochastic simulation algorithm, the next sub-volume method (NSM), which is an outgrowth of the Gillespie algorithm. We simulated a discrete event model of a Ca wave on an un-branched apical dendrite of a hippocampal pyramidal neuron on both NTW and NTW-MT. The model we used was derived from a deterministic model (1). NTW-MT scaled well with the size of the model and its execution time scaled well with the number of cores. It performed much better then the process-based NTW as well as a multi-threaded version of NTW which did not employ its multi-level priority queue. While our results indicated that execution time scaled well with the number of processors, the number of rollbacks also increased and caused the decrease in execution time to flatten. Both dynamic window management and dynamic load balancing are necessary in order to contain the number of rollbacks. The window size controls the optimism of Time Warp, preventing an excessive number of rollbacks. We have previously developed AI based algorithms for dynamic load balancing and window management-we are going to implement them in NTW-MT (simulated annealing, multi-state Q-learning and genetic algorithms) shortly. (1) S. A. Neymotin, R. A. McDougal, M. A. Sherif, C. P. Fall, M. L. Hines, and W. W. Lytton. Neuronal calcium wave propagation varies with changes in endoplasmic reticulum parameters: A computer model. Neural Computation, 27(4):898[[unable to display character: –]]924, Mar. 2015.