HFSS15: Selecting an Optimal Configuration for Distributed Analysis

With the distributed solve option (DSO) it is possible to assign more than one machine in the DSO setup panel to a single computer - assuming the computer has multiple cores. For example, a quad core computer can be listed 4 times in the DSO machine list and each CPU will run a separate simulation in the sweep. This raises the question of how to configure a distributed solution option (DSO) in combination with the multiprocessing option to take optimal advantage of the available hardware. The simplest answer is that for the overwhelming majority of situations the speed improvement will be greater if additional machines are added to a DSO than if the same CPU’s are added as multiple processors. That simple answer assumes that there is enough memory for the DSO simulations.

For multiple DSO simulations on a single machine the total memory needed is the sum of the memory used by each simulation. For example, assume we are running discrete frequency sweep and each frequency point needs 3.5GB. Our computer has only 8GB of RAM but is a quad core system. To keep the computer from going heavily into swap, which is highly inefficient, we would only want to assign this computer twice to the list of machines in the DSO setup. To take advantage of the remaining 2 CPU’s then it would make sense to set the number of distributed processors to 2. (HFSS Options: Solver Tab). This would be the optimum for this setup. In addition one would need to be sure that the amount of disk space available is also sufficient to fit all the requested simulations - typically hard drive space is not the limiting factor.