Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.705036
Title: Group-based parallel multi-scheduling methods for grid computing
Author: Abraham, G. T.
ISNI:       0000 0004 6058 4204
Awarding Body: Coventry University
Current Institution: Coventry University
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
With the advent in multicore computers, the scheduling of Grid jobs can be made more effective if scaled to fully utilize the underlying hardware and parallelized to benefit from the exploitation of multicores. The fact that sequential algorithms do not scale with multicore systems nor benefit from parallelism remains a major challenge to scheduling in the Grid. As multicore systems become ever more pervasive in our computing lives, over reliance on such systems for passive parallelism does not offer the best option in harnessing the benefits of their multiprocessors for Grid scheduling. An explicit means of exploiting parallelism for Grid scheduling is required. The Group-based Parallel Multi-scheduler for Grid introduced in this work is aimed at effectively exploiting the benefits of multicore systems for Grid job scheduling by splitting jobs and machines into paired groups and independently multi-scheduling jobs in parallel from the groups. The Priority method splits jobs into four priority groups based on job attributes and uses two methods (SimTog and EvenDist) methods to group machines. Then the scheduling is carried out using the MinMin algorithm within the discrete group pairs. The Priority method was implemented and compared with the MinMin scheduling algorithm without grouping (named ordinary MinMin in this research). The analysis of results compared against the ordinary MinMin shows substantial improvement in speedup and gains in scheduling efficiency. In addition, the Execution Time Balanced (ETB) and Execution Time Sorted then Balanced (ETSB) methods were also implemented to group jobs in order to improve on some deficiencies found with the Priority method. The two methods used the same machine grouping methods as used with the Priority method, but were able to vary the number of groups and equally exploited different means of grouping jobs to ensure equitability of jobs in groups. The MinMin Grid scheduling algorithm was then executed independently within the discrete group pairs. Results and analysis shows that the ETB and ETSB methods gain still further improvement over MinMin compared to the Priority method. The conclusion is reached that grouping jobs and machines before scheduling improves the scheduling efficiency significantly.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.705036  DOI: Not available
Keywords: Computational grids (Computer systems) ; Parallel scheduling (Computer scheduling)
Share: