Distributed computation models for aggregating a dynamic set of machines to solve complex electromagnetic problems

Summary form only given. The solution of real world electromagnetic (EM) problems, appearing in diverse technology areas, such as antenna modeling, MMIC design, EMC/EMI phenomena etc., requires, in principle, significant computing resources. The choice of the employed computational electromagnetics (CEM) method and, in particular, its scaling with the examined problem electrical size, is of major importance. We are concerned in this paper with infrastructures to solve a class of EM structures of practical interest, which are complex in terms of their EM properties, electrical dimensions and geometrical characteristics and, moreover, we are interested in determining the resonance properties of such structures (e.g. Fabry-Perot type resonators, dielectric resonators, chiral resonators). The employed CEM method is an entire domain method of moments (MoM) technique. We describe a platform for distributed computations that can accommodate a dynamic model. The architecture rests on CORBA or Java RMI for distributed message passing and relies not on code that is statically and beforehand installed on user premises, but on code that is automatically downloaded to the user's execution space. More specifically, we are using the Java applet model for client-side computations. We, then, extend this implementation by allowing the applets to host more volatile and possibly differing partial solvers. We are using the approach pioneered by the Aglets mobile agent platform, whereby an applet could in fact be a lightweight mobile agent execution environment. That is, in this approach, the applet by itself performs no useful computation, but can host any number of threads corresponding to arbitrary code segments. These code segments take the form of lightweight mobile agents and can move back and forth between the central machine and the peripheral ones, instead of the 'download-once-stay-there' applet model. This approach gives even greater flexibility, because the execution machine can decide at runtime with which agents (e.g. computation tasks) to populate a given applet. There is no other technology that can allow dynamic downloading of code of this manner to be implemented.