Solving problems on concurrent processors. Vol. 1: General techniques and regular problems
Solving problems on concurrent processors. Vol. 1: General techniques and regular problems
The cache performance and optimizations of blocked algorithms
ASPLOS IV Proceedings of the fourth international conference on Architectural support for programming languages and operating systems
Parallel programming: techniques and applications using networked workstations and parallel computers
The grid: blueprint for a new computing infrastructure
The grid: blueprint for a new computing infrastructure
Matrix Multiplication on Heterogeneous Platforms
IEEE Transactions on Parallel and Distributed Systems
High Performance Cluster Computing: Architectures and Systems
High Performance Cluster Computing: Architectures and Systems
Using MPI-2: Advanced Features of the Message Passing Interface
Using MPI-2: Advanced Features of the Message Passing Interface
Heterogeneous Networks of Workstations and the Parallel Matrix Multiplication
Proceedings of the 8th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
A cellular computer to implement the kalman filter algorithm
A cellular computer to implement the kalman filter algorithm
Efficient execution of scientific computation on geographically distributed clusters
PARA'04 Proceedings of the 7th international conference on Applied Parallel Computing: state of the Art in Scientific Computing
| Hi-index | 0.00 |
To achieve high performance, an efficient task scheduling and load balance model is necessary, our goal is to optimize the execution of parallel algorithms over a Collection of geographically separated Heterogeneous Clusters of Workstations (CoHNOWS). Some architectures are tested and an adaptive Master/Worker model is proposed, where Cluster's workers are grouped by a sub-master that acts as a single point of control, and a separate machine is used to manage inter-cluster communications over WAN. Due to network latency unpredictability, an algorithm with dynamic data distribution is used, adapting each cluster's data load. The chosen benchmark algorithm was the Matrix Multiplication, because of its scalability facilities, and the MPI communications library for its portability. The testbed system used is composed of two heterogeneous dedicated clusters (HNOW), one in Spain and the other in Brazil, using a nondedicated WAN (Internet) as the interconnection network between them.