Friday, November 14, 2008

Integration of solvers using Neurospaces

subject: a soliton solver and a compartmental solver


Just wanted to clarify something that was brought up yesterday during our journalclub.

So what I said was that it is a bad idea to integrate a soliton solver (fluid dynamics, hyperbolic partial differential equations) with a compartmental solver (parabolic partial differential equations), because the structure of the equations and their solutions is very different.

This does not mean that integration with neurospaces / G3 is a bad idea. The only thing to pay attention to, is that one has to keep the code base of the two solvers entirely separate, for ease of development / maintenance / testing. During simulation, both are coupled by an appropriate communication framework that comes with neurospaces / G3.

Such a situation is very close to the TMS stimulation engine that Reese and I have worked on.

This is very different from hsolve of G2 where all kinds of solvers and their internal communication mechanisms share a single entangled code base (and development ceased).

So again, it is just a matter of modular design,

Ah and BTW. before I forget to mention: this also facilitates optimization. I recently added a very simple optimizer to the SSP scheduler, and it runs about 30% faster now. Single neuron simulations on G3 run faster than on G2.

This will be posted on the neurospaces blog.

No comments: