Parallel Computing Research The following are the primary research projects of the Parallel Computing Laboratory. The diagram on the left illustrates how of how some of these projects interrelate. Major Research Areas: Parallel Simulation Parallel Languages This research is being supported under the following contracts. iMASH - Interactive Mobile Application Support for Heterogeneous clients DOMAINS - Design of Mobile Adaptive Networks Using Simulation and Agent Technology AASERT - Science and Engineering Research Training POEMS - Performance Modeling of Parallel Computer/Communications Systems TRAVLER - Parallel Network Simulation SESAME - System Software Measurement and Evaluation Parallel Simulation Our parallel simulation research is in the areas of languages and protoocls for parallel simulation and in their application to simulation of VLSI designs, networks, and parallel programs. The languages designed and implemented by our group include Parsec, Maisie, MOOSE, and COMPOSE. Parsec is a C-based simulation language that can support a variety of simulation protocols including a sequential global event-list algorithm, parallel conservative algorithms based on null messages, conditional events, and a new conservative protocol that combines null messages with conditional events, and a parallel optimistic algorithm. Parsec and Maisie have been implemented on a 24-node IBM SP, an 8-processor Sparc1000, and a network of workstations. MOOSE is an object-oriented extension of Maisie, and COMPOSE is a C++ library that supports parallel execution of simulation models written in C++ using a conservative, optimistic, or mixed simulation protocol. COMPOSE has been implemented on the IBM SP distributed memory multicomputer. The simulation protocols designed by our group include ADAPT and MIDAS. The ADAPT simulation algorithm supports parallel execution of a model such that different sub-models may be synchronized using either conservative or optimistic protocols, and the synchronization mechanism can be changed dynamically. MIDAS supports the execution of parallel hybrid models. A parallel hybrid model includes subsystems that are operational parallel software that use physical processor clocks and also include simulation models of other subsystems. The languages and protocols have been used to design parallel implementations of a number of models including gate and switch-level models of VLSI designs, data parallel programs implemented on distributed memory multicomputers, and wireless and ATM networks. Parallel Languages The parallel environments designed and implemented by our group include UC, a C-based language that includes both data and task parallel programming paradigms, Parsec, a C-based parallel simulation language, and COMPOSE, a C++ library for parallel discrete-event simulation. UC implementations are available for the Connection Machine CM-2, a network of workstations and for distributed memory multicomputers. A simulator that predicts the performance of UC programs on parallel platforms is also available.
	Last updated Friday, 04-Aug-2000 12:13:47 PDT