parallel computation in linux cluster

It will be useful to ensure each run is optimally-coded: make sure things that can be reused are being reused, that big arrays are pre-built, not grown (which is slow), picking the fastest optimizer for your particular problem, good start values.

Other than that, you need more cores. Find a large processor farm - several leading universities have high-performance facilities with 2000 or more 8-32 core mini-computer level cores. Or setup an cloud service (Amazon and google both sell these). You can recruit as many machines as you like.

With setups like these, the task should run over-night or quicker.

Hi Veronica

We have had good experiences with cloud computing services, and are preparing a 'how-to' article for publication. Obviously, one would want to try out a small number of simulations locally first, then the same scale on the cloud, before burning money on a possibly errant set of analyses. If you have no such funds available for this type of activity, then the local cluster may be best. Possibly, relatively few threads (cores), but a larger number of separate jobs, would be more efficient than enabling a large number of threads. There is something of a law of diminishing returns with parallel computation, at least in my experience with OpenMx and multicore analyses.