Abstract
In the quest of peta- and exa-scale computing, researchers have been interested in integrating optical communication technologies into parallel computing models. Optical bus parallel computing models is one type of recent model that has gained much interest in the past decade. Over ten such models have been developed; many of which share significant similarities. These models are mostly theoretical parallel computation models, and only limited systems related research has been reported. The Linear Array with a Reconfigurable Pipelined Bus System (LARPBS), 1996, is one of the more popular models in the literature. This research addresses the lack of systems related work inherent in the LARPBS model. The research is also applicable to some of the other models as well. This talk describes some recent work on the systems and implementation aspects of LARPBS, specifically: (a) the newly proposed LARPBS(p) model that reflects architectural parameterization of the optical bus parallel model, (b) simulations of communication traffic induced by SIMD and MIMD type algorithms together with collision avoidance conditions, (c) an optical power budget, and (d) extensions to free-space optical parallel models. These results have pushed the field to a point where system simulation and prototyping, prelude to hardware implementations, are the next step. This talk concludes with the conjecture that peta- and exa-scale computing could be enabled by optical bus parallel computing models, but only after further research that combines systems and theory.