Results 1  10
of
10
Portable and Architecture Independent Parallel Performance Tuning Using BSP
 PARALLEL COMPUTING
, 2002
"... A callgraph profiling tool has been designed and implemented to analyse the efficiency of programs written in BSPlib. This tool highlights computation and communication imbalance in parallel programs, exposing portions of program code which are amenable to improvement. A unique feature of this prof ..."
Abstract

Cited by 20 (5 self)
 Add to MetaCart
A callgraph profiling tool has been designed and implemented to analyse the efficiency of programs written in BSPlib. This tool highlights computation and communication imbalance in parallel programs, exposing portions of program code which are amenable to improvement. A unique feature of this profiler is that it uses the BSP cost model, thus providing a mechanism for portable and architectureindependent parallel performance tuning. In order to test the capabilities of the model on a realworld example, the performance characteristics of an SQL query processing application are investigated on a number of different parallel architectures.
Collection Types for Database Programming in the BSP Model
 In Fifth Euromicro Workshop on Parallel and Distributed Processing. IEEE CS
, 1997
"... We study the pragmatics of integrating collection types, that model a broad class of nonnumerical applications, into the Bulk Synchronous Parallel (BSP) model which abstracts a diversity of parallel architectures using just four numerical parameters. We outline how the collection types have been bui ..."
Abstract

Cited by 6 (2 self)
 Add to MetaCart
(Show Context)
We study the pragmatics of integrating collection types, that model a broad class of nonnumerical applications, into the Bulk Synchronous Parallel (BSP) model which abstracts a diversity of parallel architectures using just four numerical parameters. We outline how the collection types have been built ontop of the direct BSP programming environment provided by BSPlib, give results on a SGI PowerChallenge and IBM SP2, and discuss how these types can help implement object databases. 1 Introduction Parallel computers have been successfully deployed in many scientific and numerical application areas, but their use in commercial applications, which are often nonnumerical in nature, has been scarce. One of the impediments in the longterm commercial uptake of parallel computing has been the proliferation of differing machine architectures and corresponding programming models. However, due to several technological and economic reasons, the various classes of parallel computers such as shar...
PRAM Programming: Theory vs. Practice
 IN PROCEEDINGS OF 6TH EUROMICRO WORKSHOP ON PARALLEL AND DISTRIBUTED PROCESSING
, 1997
"... In this paper we investigate the practical viability of PRAM programming within the BSP framework. We argue that there is a necessity for PRAM computations in situations where the problem exhibits poor data locality. We introduce a C++ PRAM simulator that is built on top of the Oxford BSP Toolset, B ..."
Abstract

Cited by 5 (1 self)
 Add to MetaCart
In this paper we investigate the practical viability of PRAM programming within the BSP framework. We argue that there is a necessity for PRAM computations in situations where the problem exhibits poor data locality. We introduce a C++ PRAM simulator that is built on top of the Oxford BSP Toolset, BSPlib, and provide a succinct PRAM language. Our approach achieves simplicity of programming over directmode BSP programming for reasonable overhead cost. We objectively compare optimised BSP algorithms with PRAM algorithms implemented with our C++ PRAM library and provide encouraging experimental results for the latter style of programming.
G.: Design of the force field task assignment method and associated performance evaluation for desktop grids
 In: Proceedings of the 2005 Fourth International Conference on Grid and Cooperative Computing, SpringerVerlag
, 2005
"... Abstract. In the case of desktop grids, a single hardwaredetermined latency and constant bandwidth between processors cannot be assumed without incurring in unnecessary error. The actual network topology is determined not only by the physical hardware, but also by the instantaneous bandwidth availa ..."
Abstract

Cited by 1 (1 self)
 Add to MetaCart
(Show Context)
Abstract. In the case of desktop grids, a single hardwaredetermined latency and constant bandwidth between processors cannot be assumed without incurring in unnecessary error. The actual network topology is determined not only by the physical hardware, but also by the instantaneous bandwidth availability for parallel processes to communicate. In this paper we present a novel task assignment scheme which takes the dynamic network topology into consideration along with the traditionally evaluated variables such as processor availability and potential. The method performs increasingly better as the grid size increases.
Optimistic Concurrency Control for Inverted Files in Text Databases
"... Inverted files are frequently used as index data structures for very large text databases. Most applications of this data structure are for readonly query operations. However, the problem of introducing update operations has deserved little attention so far and yet it has important applications. In ..."
Abstract
 Add to MetaCart
Inverted files are frequently used as index data structures for very large text databases. Most applications of this data structure are for readonly query operations. However, the problem of introducing update operations has deserved little attention so far and yet it has important applications. In this paper we propose an optimistic concurrency control algorithm devised to handle mixes of update operations and readonly queries efficiently. We work on top of the BSP model of parallel computing and take advantage of a BSP Time Warp realization to formulate the proposed algorithm. We present a comparison with the traditional lockbased approach and the results show that optimism is particularly efficient in this case.
On BulkSynchronous DistributedMemory Parallel Processing of RelationalDatabase Transactions
"... This paper describes two parallel algorithms for the ecient processing of relational database transactions and presents a performance analysis of them. These algorithms are built upon the bulksynchronous parallel model of computation. The wellde ned structure of this model enabled us to evalua ..."
Abstract
 Add to MetaCart
This paper describes two parallel algorithms for the ecient processing of relational database transactions and presents a performance analysis of them. These algorithms are built upon the bulksynchronous parallel model of computation. The wellde ned structure of this model enabled us to evaluate their performance by using an implementation independent and yet empirical approach which includes the eects of synchronization, communication and computation.
Verifying Functional Bulk Synchronous Parallel Programs Using the Coq System
, 2003
"... The Bulk Synchronous Parallel ML (BSML) is a functional language for Bulk Synchronous Parallel (BSP) programming. It is based on an extension of the #calculus by parallel operations on a parallel data structure named parallel vector, which is given by intention. We present the formal proofs of c ..."
Abstract
 Add to MetaCart
(Show Context)
The Bulk Synchronous Parallel ML (BSML) is a functional language for Bulk Synchronous Parallel (BSP) programming. It is based on an extension of the #calculus by parallel operations on a parallel data structure named parallel vector, which is given by intention. We present the formal proofs of correctness of BSML programs in the Coq proof assistant. Such development demonstrates the usefulness of higherorder logic in the process of software certification and parallel applications. They also show that proof of rather complex parallel algorithms may be made with inductive types by using the certified programs.
Contents
, 2004
"... 1.1 The BSP Model............................................ 4 1.2 Explicit processes and the SPMD programming style....................... 5 1.3 Overview of the core BSMLlib library................................ 6 ..."
Abstract
 Add to MetaCart
(Show Context)
1.1 The BSP Model............................................ 4 1.2 Explicit processes and the SPMD programming style....................... 5 1.3 Overview of the core BSMLlib library................................ 6
Division de Informatica
"... This paper describes two parallel algorithms for the eÆcient processing of relational database transactions and presents a performance analysis of them. These algorithms are built upon the bulksynchronous parallel model of computation. The welldened structure of this model enabled us to evaluate t ..."
Abstract
 Add to MetaCart
(Show Context)
This paper describes two parallel algorithms for the eÆcient processing of relational database transactions and presents a performance analysis of them. These algorithms are built upon the bulksynchronous parallel model of computation. The welldened structure of this model enabled us to evaluate their performance by using an implementation independent and yet empirical approach which includes the eects of synchronization, communication and computation. The analysis reveals that the algorithm which borrows ideas from optimistic parallel discrete event simulation achieves better performance than the classical approach for synchronizing concurrent transactions on a distributed memory system. 1