Speedup possible?

[neuroml]

Speedup possible?

[pgleeson]

@lungd Bear in mind that what jNeuroML is doing there is spawning an external process (Neuron), running it and waiting until it completes. All of the simulation of the worm takes place in Neuron and this should be the bulk of the time of the whole process.

In some small networks it may be quicker to run the simulation in jNeuroML itself, it's inherently slower (and normally required a smaller dt, e.g. 0.01 ms) but you don't have the overhead of writing Neuron py/hoc/mod files, compiling the mod file and launching an external process.

One way or another you should experiment with adjusting the dt value, as this is the biggest factor in determining the run time of simulations. If you are running many simulations, you might get away with 0.1 ms in Neuron, but always check the behaviour against runs at 0.01 in nrn (or 0.001 in jnml natively) to make sure the overall behavour is captured.

[lungd]

NeuronWriter creates a file with a content containing snippets with following structure:

f = open(...)
num_points = len(py_v_time)

for i in range(num_points):
    f.write(...)
f.close()

A possible improvement would be to create a string first and call write only once, e.g:

num_points = len(py_v_time)
file_string = ''.join(... for i in range(len(py_v_time)))
with open(...) as f:
    f.write(file_string)