A decentralized convergence detection algorithm for asynchronous parallel iterative algorithms

Bahi, Jacques M.; Contassot-Vivier, Sylvain; Couturier, Raphaël; Vernier, Flavien

doi:10.1109/tpds.2005.2

Cited by 40 publications

(33 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Convergence can be detected using a centralized algorithm described in [2] or a decentralized one, that is a more general version, as described in [3].…”

Section: Convergence Detectionmentioning

confidence: 99%

Solving large sparse linear systems in a grid environment: the GREMLINS code versus the PETSc library

2011

View full text Add to dashboard Cite

International audienceSolving large sparse linear systems is essential in numerous scientific domains. Several algorithms, based on direct or iterative methods, have been developed for parallel architectures. On distributed grids consisting of processors located in distant geographical sites, their performance may be unsatisfactory because they suffer from too many synchronizations and communications. The GREMLINS code has been developed for solving large sparse linear systems on distributed grids. It implements the multisplitting method that consists in splitting the original linear system into several subsystems that can be solved independently. In this paper, the performance of the GREMLINS code obtained with several libraries for solving the linear subsystems is analyzed. Its performance is also compared with that of the widely used PETSc library that enables one to develop portable parallel applications. Numerical experiments have been carried out both on local clusters and on distributed grids

show abstract

“…Convergence can be detected using a centralized algorithm described in [2] or a decentralized one, that is a more general version, as described in [3].…”

Section: Convergence Detectionmentioning

confidence: 99%

Solving large sparse linear systems in a grid environment: the GREMLINS code versus the PETSc library

2011

View full text Add to dashboard Cite

show abstract

“…On the other hand, if the preconditioning iteration is performed on dedicated but geographically separated hardware such as the DAS-3 architecture, some sophisticated decentralised convergence detection algorithm may have to be employed [55]. In analogy to the aforementioned case, determining how accurate one should solve the preconditioning iteration is far from trivial.…”

Section: Convergence Detectionmentioning

confidence: 99%

Parallel Scientific Computing on Loosely Coupled Networks of Computers

Collignon

Gijzen

2009

Lecture Notes in Computational Science and Engineering

View full text Add to dashboard Cite

Efficiently solving large sparse linear systems on loosely coupled networks of computers is a rich and vibrant field of research. The induced heterogeneity and volatile nature of the aggregated computational resources present numerous algorithmic challenges. Designing efficient numerical algorithms for said systems is a complex process that brings together many different scientific disciplines. This book chapter is divided into two distinct parts. The purpose of the first half (Sect. §2- §4) is to give a bird's view of the issues pertaining to designing efficient numerical algorithms for Grid computing. It kicks off by clearly stating the problem and exposing the various bottlenecks, subsequently followed by the presentation of potential solutions. Thus, the stage is set and Sect. §3 proceeds by detailing classical iterative solution methods, along with the concept of asynchronism, which is a highly favorable quality in the context of Grid computing. The first half is wrapped up by explaining how asynchronism can be introduced into faster but more complicated subspace methods. The general idea is that by using an asynchronous method as a preconditioner, the best of both worlds can be combined. The advantages and disadvantages of this approach are discussed in minute detail. The second half (Sect. §5) contains discussions on the various intricacies related to implementing the proposed algorithm on Grid computers. Section §6 gives some concluding remarks along with suggestions for further reading.

show abstract

“…Note that while there has been significant work on parallel asynchronous iterative computations in recent years, these efforts have focused on algorithmic and implementation issues such as convergence rate, termination detection, and load balancing [11,12,[15][16][17]. The research presented in this paper leverages these efforts and focuses on the development and execution of applications based on these algorithms on Desktop Grid environments with Internet-scale connectivity.…”

Section: Parallel Asynchronous Iterative Algorithms and Applicationsmentioning

confidence: 99%

“…CometG provides single master mode as well as multiple master mode. In multiple master mode, hierarchical or decentralized termination algorithms [12] are supported based on the organization of the masters. A master module has five components:…”

Section: The Application Layermentioning

confidence: 99%

A Decentralized Computational Infrastructure for Grid-Based Parallel Asynchronous Iterative Applications

Parashar

2006

J Grid Computing

View full text Add to dashboard Cite

Parallel asynchronous iterative algorithms relax synchronization and communication requirements, and can potentially extend Desktop Grids beyond embarrassingly parallel applications to support a broader class of parallel iterative applications. This paper presents the design and implementation of CometG, a decentralized (peerto-peer) computational infrastructure that extends Desktop Grid environments to support these applications. CometG provides a decentralized and scalable tuple space, efficient communication and coordination support, and application-level abstractions that can be used to implement Desktop Grid applications based on parallel asynchronous iterative algorithms using the master-worker/BOT paradigm. The deployment and evaluations of CometG and a CometG-based application in a wide-area environment using the PlanetLab [7] test bed, as well as a campus network are presented.Key words decentralized (peer-to-peer) tuple space . Desktop Grids . parallel asynchronous iterative algorithms Abbreviations BOT bag of task PDE partial differential equation

show abstract

A decentralized convergence detection algorithm for asynchronous parallel iterative algorithms

Cited by 40 publications

References 16 publications

Solving large sparse linear systems in a grid environment: the GREMLINS code versus the PETSc library

Solving large sparse linear systems in a grid environment: the GREMLINS code versus the PETSc library

Parallel Scientific Computing on Loosely Coupled Networks of Computers

A Decentralized Computational Infrastructure for Grid-Based Parallel Asynchronous Iterative Applications

Contact Info

Product

Resources

About