high-performance heterogeneous computing || part introduction

1
PART IV APPLICATIONS In Chapter 7, we have noted that the design of heterogeneous parallel algo- rithms significantly outdistances their implementation in the form of portable and efficient applications. Most of the successful applications for large-scale heterogeneous systems, such as seti@home, use a trivial model of parallelism, where the whole problem is partitioned into a huge number of fully indepen- dent tasks that can be processed in parallel. The design and implementation of tightly coupled high-performance computing applications for heteroge- neous platforms is definitely an underdeveloped area. At the same time, first such applications, based on nontrivial models and algorithms, have started appearing. In this part, we present some recent results in this area. Chapter 9 introduces Heterogeneous PBLAS, a set of parallel linear algebra subpro- grams for heterogeneous computational clusters. Parallel processing of remotely sensed hyperspectral images on heterogeneous clusters is presented in Chapter 10. Both applications are implemented using HeteroMPI, the extension of MPI for heterogeneous parallel computing introduced in Chapter 8. An astrophysical application that simulates the evolution of clusters of gal- axies in the universe on a heterogeneous computational grid is described in Chapter 11. Its implementation in GridSolve and SmartGridSolve are dis- cussed and compared. 169 High-Performance Heterogeneous Computing, by Alexey L. Lastovetsky and Jack J. Dongarra Copyright © 2009 John Wiley & Sons, Inc.

Upload: jack-j

Post on 06-Jun-2016

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: High-Performance Heterogeneous Computing || Part Introduction

PART IV

APPLICATIONS

In Chapter 7 , we have noted that the design of heterogeneous parallel algo-rithms signifi cantly outdistances their implementation in the form of portable and effi cient applications. Most of the successful applications for large - scale heterogeneous systems, such as seti@home, use a trivial model of parallelism, where the whole problem is partitioned into a huge number of fully indepen-dent tasks that can be processed in parallel. The design and implementation of tightly coupled high - performance computing applications for heteroge-neous platforms is defi nitely an underdeveloped area. At the same time, fi rst such applications, based on nontrivial models and algorithms, have started appearing. In this part, we present some recent results in this area. Chapter 9 introduces Heterogeneous PBLAS, a set of parallel linear algebra subpro-grams for heterogeneous computational clusters. Parallel processing of remotely sensed hyperspectral images on heterogeneous clusters is presented in Chapter 10 . Both applications are implemented using HeteroMPI , the extension of MPI for heterogeneous parallel computing introduced in Chapter 8 . An astrophysical application that simulates the evolution of clusters of gal-axies in the universe on a heterogeneous computational grid is described in Chapter 11 . Its implementation in GridSolve and SmartGridSolve are dis-cussed and compared.

169

High-Performance Heterogeneous Computing, by Alexey L. Lastovetsky and Jack J. DongarraCopyright © 2009 John Wiley & Sons, Inc.