Fairness Categorization Policy Of Queuing Theory For Geographic Information System Job Scheduling

Geographic Information System (GIS) is a compute-intensive plus data-intensive application that deals with substantial amount of spatial data processing and rendering of three-dimensional (3D) images of the locations. Besides research work on data or image processing part of GIS applications, schedu...

Full description

Saved in:
Bibliographic Details
Main Author: Kheoh, Hooi Leng
Format: Thesis
Language:English
Published: 2013
Subjects:
Online Access:http://eprints.usm.my/43533/1/Kheoh%20Hooi%20Leng24.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Geographic Information System (GIS) is a compute-intensive plus data-intensive application that deals with substantial amount of spatial data processing and rendering of three-dimensional (3D) images of the locations. Besides research work on data or image processing part of GIS applications, scheduling of GIS workload can be further studied to improve the performance of GIS applications. In this regards, this thesis proposes an algorithm of job scheduler named Fair Categorized Queue Scheduling (FCQS) which distributes jobs of GIS applications efficiently. Queuing theory is applied in FCQS for job scheduling processes meanwhile the GIS job arrivals are distributed according to Poisson distribution. Each category of jobs is served along with First-Come First-Served (FCFS) basic using the Multiple Queues Multiple Machines (MQMM) configuration. The experiment through simulation has been carried out to evaluate the performance of FCQS and other queue configurations such as Single Queue Single / Multiple Machine(s) (SQSM / SQMM) and Multiple Queues Single / Multiple Machines(s) (MQSM / MQMM). The results proved that the FCQS algorithm achieved the highest throughput with 24 jobs or 72.727% more than the lowest throughput of SQSM. Additionally, the total Input / Output (IO) transferring time can be reduced up to 49.194% by using multiple jobs processing compared to single job processing within small jobs, attaining lower average turnaround time and waiting time simultaneously. Last but not least, the optimization of grid resources has been significantly improved by decreasing total pending jobs to 28.261% instead of the highest 52.174%.