Showing papers on "Degree of parallelism published in 1984"

Queue-based multi-processing LISP

Abstract: As the need for high-speed computers increases, the need for multi-processors will be become more apparent. One of the major stumbling blocks to the development of useful multi-processors has been the lack of a good multi-processing language—one which is both powerful and understandable to programmers.Among the most compute-intensive programs are artificial intelligence (AI) programs, and researchers hope that the potential degree of parallelism in AI programs is higher than in many other applications. In this paper we propose multi-processing extensions to Lisp. Unlike other proposed multi-processing Lisps, this one provides only a few very powerful and intuitive primitives rather than a number of parallel variants of familiar constructs.

Materials for optical information processing.

Abstract: The goal for optical information processing is to use the unique characteristics of light which are not readily achieved with electronic devices, namely, ultrahigh speed (picoseconds), a high degree of parallelism (image processing), and conductor-free interconnection. The requirements of the nonlinear materials to perform such functions, using all-optical interactions, are discussed and the limitations of the nonlinear mechanisms are outlined.

Highly parallel inference engine PIE —Goal rewriting model and machine architecture—

Abstract: Logic programming is expected to make knowledge information processing feasible. However, conventional Prolog systems lack both processing power and flexibility for solving large problems. To overcome these limitations, an approach is developed in which natural execution features of logic programs can be represented using Proof Diagrams. AND/ OR parallel processing based on a goal-rewriting model is examined. Then the abstract architecture of a highly parallel inference engine (PIE) is described. PIE makes it possible to achieve logic/control separation in machine architecture. The architecture proposed here is discussed from the viewpoint of its high degree of parallelism and flexibility in problem solving in comparison with other approaches.

Division and sign detection algorithms for residue number systems

Abstract: This paper is concerned with the operations of division and sign detection in residue number systems. CORDIC, a well-known iterative algorithm originally designed for conventional floating point representation, is adapted for residue division. Since sign detection is required by the division algorithm we have also considered this problem. A modified sign detection algorithm which yields a high degree of parallelism is introduced resulting in time efficient division and sign detection.

Towards automated design of multicomputer system for real-time applications (architecture, task division)

Abstract: With the recent advances in VLSI technology it has now become feasible to design and implement computer systems consisting of a large number of processors. The suitability of such systems for specific applications depends primarily upon how well the system architecture corresponds to the structure of the algorithm(s) to be implemented. This work aims at the development of a software tool for designing an application-oriented or application-driven inhomogeneous multicomputer system (MCS) by evaluating the performance of the candidate architectures with respect to the algorithmic requirements of the application (task). The structure and the run-time ordering (data dependency of subtasks) of the applications are modeled by computation flow graphs (CFGs). The operating environments of the various MCSs have been represented by computing resource graphs (CRGs). CFG represents the degree of parallelism present in the task and the amount of computation and communication at each of its subtasks. Real-time applications in which continuous streams of input data are to be analyzed are considered in this work. An O(N('2)) time heuristic static allocation algorithm is presented to map the CFG to some CRG. The allocation algorithm provides a unique approach for reducing the interprocessor communication caused by the assignment of two communicating subtasks to different processors. This particular scheduling problem is shown to be NP-complete, and the heuristic algorithm presented in the thesis is shown to be bounded by a simple function of the number of levels in the CFG and the degree of inhomogeneity in the CRG. Performance of MCS for some application is evaluated in terms of speed-up, turn around time, resource utilization, and cardinality of the mapping, with the objective of evaluating an optimization function which could provide guidelines for selecting an architecture. The usefulness of the approach is demonstrated by comparing various MCSs for applications such as dynamic scene analysis problems, weather forecasting problems, fast fourier transforms, etc. The performance of the multicomputer systems for real-time applications is shown to be greatly influenced by its architecture. The software for the allocation algorithm and the simulation of MCS has been implemented on VAX 11/780 4.2 BSD UNIX using PASCAL and C. Finally, extensions of the project as a CAD tool and directions for the further research are outlined.

A Strategy for Decomposing Complex Queries in a Heterogeneous DDB

Abstract: In a generalised distributed database system with decentralised controls and heterogeneous and pre-existing nodes, queries can be very complex, particularly if they provide a data integration facility. We describe here an algorithm for the optimal decomposition of such queries into subqueries, taking into consideration the availability of nodal operations (some nodes may not be able to perform all operations) and other factors. This algorithm is being implemented in the PRECI* system. In a distributed database system, an efficient query processing strategy is essential for ameliorated performance. In general there will be many possible strategies for processing a particular query, and ideally each of these should be evaluated in order to determine the best strategy. Unfortunately, however, the,problem of selecting optimal strategies for complex queries 1s NP-complete, so it is not feasible to evaluate every strategy for such queries. Many query decomposition algorithms have therefore been designed to produce optimal or near-optimal strategies only under a set of highly restrictive assumptions that apply to a particular implementation (1, 2). The only algorithm developed for a DDB which allows heterogeneous pre-existing databases as nodes is that of the MULTIBASE project (3). They Permission 10 copy without fee all or part of this material is gmnted provided that the copies are not made or distributed for direct commercial advantage, the VLDB copyright notice and the title of the publication and its date appear, and notice Zr given that copying is by pennksion of the Very Lave Data Lkse Endowment. To copy otherwise, or to repubhh, requbes a fee and/or special permission from the Endowment. Proceedings of the Tenth Intematlonal Conterence on Very Large Dats Bsses. evaluate strategies in terms of data movement and local processing, disregarding response time. All final processing is done at the result node, so that there is little parallel processing. In PRECI* (4, 6) we have aimed to achieve a higher degree of parallelism while still taking account of data movement and local processing. Recognising the NP completeness of the general problem, we propose to tackle it in two stages: optimal decomposition of a query into subqueries and the subsequent allocation of the subqueries optimally to nodes, taking into account the presence of replicated data and network characteristics. We believe that for complex queries in heterogeneous DDBs, this two-staged approach is most fruitful although it may not yield the optimal strategy. In this paper, we present Only the decomposition strategy which involves determining the order of operations to be performed, applying transformations to the original query expression in order to reduce the total cost (in terms of data movement and local processing) or to reduce response time by increasing parallelism without increasing the total cost. Operations are then grouped into subqueries. For the node allocation stage, we decide where each subquery should be executed, evaluating each possible strategy according to the total cost and response time. There may be many possible strategies since, in PRECI*, operations on external data (data sent from another node) may be performed at any node which supports an appropriate interface, unlike in MULTIBASE where only the result node is used for such operations. We assume that a given node may not be able to perform all PAL operations, and this is taken into consideration in the query decomposition strategy. PRECI* is a generalised distributed database management system supporting heterogeneous,. possibly pre-existing, databases as nodes (4, 5). It also allows data replication under global supervision. Any database may join PRECI* as a node, provided it supplies a minimal relational interface, and any network could be used to link the nodes. Queries to the DDB are expressed in the PRECI algebraic language (PAL), Singapore, August, 1994

Parallel Algorithms — Theory and Limitations

Abstract: Whereas many mental processes are governed by a high degree of parallelism, algorithms were restricted to a sequential flow of operations and data for many decades. This limitation was strongly supported by the serial von Neumann type computer architecture. Favoured by the design and availability of computers with parallel features in several levels and domains, during the last twenty years the design, study and implementation of parallel algorithms gained an increasing interest.