Efficient scheduling of parallel applications on workstation clusters
Efficient scheduling of parallel applications on workstation clusters
In this thesis we investigate the improved scheduling of parallel applications on workstation clusters using the Message-Passing Interface (MPI) software environment. MPI is a collaboration by many organisations to define a de facto message-passing standard. The original specification has only a simple scheduling mechanism for submitting parallel applications on networks of workstations.
Demonstrators applications that are commonly employed in engineering and scientific fields have been parallelised using the MPI paradigm to provide assessment of the proposed mechanism to schedule effectively parallel applications.
A research survey has been conducted to investigate the state-of-the-art of clustering systems including the facilities provided for programmers to execute parallel applications on networks of heterogeneous workstations. In addition, load balancing techniques have been evaluated to examine how these mechanisms can be applied to the workload distribution especially on heterogeneous workstation clusters. As a result from this study, a hybrid dynamic/static workload distribution mechanism described as selective load balancing is proposed as an original contribution to improve load balancing on heterogeneous workstation clusters.
Finally, a novel lightweight approach is presented to extend the MPI specification when using networks of heterogeneous workstations which extends the existing concepts of the software environment. This approach targets uncontrolled configurations. The design philosophy of this new environment, called Selective-MPI (S-MPI), is described in detail. Enhanced results of the demonstrator applications using this system are compared with standard MPI. These results demonstrate the improved efficiency of S-MPI, with a reduction in elapsed-time of up to 38%.
University of Southampton
Dantas, Mario Antonio Ribeiro
1996
Dantas, Mario Antonio Ribeiro
Dantas, Mario Antonio Ribeiro
(1996)
Efficient scheduling of parallel applications on workstation clusters.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
In this thesis we investigate the improved scheduling of parallel applications on workstation clusters using the Message-Passing Interface (MPI) software environment. MPI is a collaboration by many organisations to define a de facto message-passing standard. The original specification has only a simple scheduling mechanism for submitting parallel applications on networks of workstations.
Demonstrators applications that are commonly employed in engineering and scientific fields have been parallelised using the MPI paradigm to provide assessment of the proposed mechanism to schedule effectively parallel applications.
A research survey has been conducted to investigate the state-of-the-art of clustering systems including the facilities provided for programmers to execute parallel applications on networks of heterogeneous workstations. In addition, load balancing techniques have been evaluated to examine how these mechanisms can be applied to the workload distribution especially on heterogeneous workstation clusters. As a result from this study, a hybrid dynamic/static workload distribution mechanism described as selective load balancing is proposed as an original contribution to improve load balancing on heterogeneous workstation clusters.
Finally, a novel lightweight approach is presented to extend the MPI specification when using networks of heterogeneous workstations which extends the existing concepts of the software environment. This approach targets uncontrolled configurations. The design philosophy of this new environment, called Selective-MPI (S-MPI), is described in detail. Enhanced results of the demonstrator applications using this system are compared with standard MPI. These results demonstrate the improved efficiency of S-MPI, with a reduction in elapsed-time of up to 38%.
This record has no associated files available for download.
More information
Published date: 1996
Identifiers
Local EPrints ID: 462959
URI: http://eprints.soton.ac.uk/id/eprint/462959
PURE UUID: 7fad1692-48af-4a38-a42e-7340b5ce21bd
Catalogue record
Date deposited: 04 Jul 2022 20:30
Last modified: 04 Jul 2022 20:30
Export record
Contributors
Author:
Mario Antonio Ribeiro Dantas
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics