Abstract not found

Portable Library Support for Irregular Applications by Chih-Po Wen Doctor of Philosophy in Computer Science University of California at Berkeley Professor Katherine A. Yelick, Chair Building portable parallel programs on distributed memory multiprocessors and workstation networks is a complex task that is greatly facilitated by powerful infrastructure. In this dissertation, we develop important components of that infrastructure, focusing on irregular applications such as unstructured mesh computations, search problems, and discrete event simulation. We use a library-based approach to building such applications. The library provides a uniform programming interface on multiple platforms and has highly tuned implementations developed by the library programmer. Therefore, applications built on the library can be portable both in functionality and in performance. We describe the major components of our parallel data structure library called Multipol, including two of the more irregular dat...

. In this paper, we describe an extension of Common Lisp which allows the definition of parallel programs within that functional and object-oriented language. In particular, the extensions are the introducing of active objects, sending synchronous and asynchronous messages between them, automatic and manual distribution of active objects to object spaces, and transparent object managing. With these extensions object-oriented parallel programming on a workstation cluster using different Common Lisp images is possible. These concepts are implemented as an extension of Allegro Common Lisp subsumed by the name NetCLOS. 1 Introduction One of the big problems of Artificial Intelligence (AI) is getting its applications answer in time. Parallel computation is one way to solve this problem. But though there are many parallel implementations of basic AI techniques, there are very few AI applications which use them. This drawback is ascertained due to two reasons: -- Most of these implem...

Discrete relaxation is frequently used to compute the fixed point of a discrete system where / is monotonic with respect to some partial order Given an appropriate initial value for X, discrete relaxation repeats the assignment until a fixed point for / is found. Monotonicity of / with respect to is a sufficient (but in general not necessary) condition for iterative, hill-climbing techniques such as discrete relaxation to find the fixed point of /. In this paper we introduce monotonic asynchronous iteration as a novel way of implementing parallel discrete relaxation in problem domains for which monotonicity is a necessary condition. This is an optimistic technique that maintains monotonicity without limiting concurrency, resulting in good parallel performance. We illustrate this technique with the parallel implementation of a constraint satisfaction system that computes globally consistent solutions, and present performance numbers for experiments on a shared-memory implementation. The performance numbers show that it is indeed possible to obtain a reasonable speedup when parallelizing global constraint satisfaction. We believe that monotonic asynchronous iteration is applicable to parallel discrete relaxation in general. 1