Batch mode heuristic approaches for efficient task scheduling in grid computing system
The concept of grid computing originated in the early 1990s as a metaphor for making computer power as easy as accessing an electric power grid. Grid computing appears to be a promising trend due to its ability to make the computational cost more cost-effective, the use of a given amount of compute...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2016
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/69359/1/FSKTM%202016%2023%20IR.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The concept of grid computing originated in the early 1990s as a metaphor for making computer power as easy as accessing an electric power grid. Grid
computing appears to be a promising trend due to its ability to make the computational cost more cost-effective, the use of a given amount of computer
resources, as a way to solve the problems that cannot be approached without an enormous amount of computing power, and its capability of utilising the resources
of many computers which are not in use for other computational tasks. For desirable use (application) of the capabilities of large distributed systems like Grid,
an efficient and effective scheduling algorithm is required for reducing total
completion time and advancement of load balancing. Many algorithms have been
implemented to solve the grid scheduling problem. These include Min-Min and Max-Min tasks scheduling algorithms, the former finds a task with minimum execution time and assigned to a resource that is able to produce it with minimum completion time, whereas the latter finds a task with maximum execution time and
assigned to a resource that is abe to produce it with minimum completion time. Min-Min task scheduling algorithm has two clear weaknesses, a high value of makespan being generated and low resource utilisation when the numbers of tasks with minimum execution time are more than the number of tasks with maximum execution time. In Max-Min algorithm, a high completion time and resource imbalance are the two issues arise when the number of tasks with maximum execution time are more than the number of tasks with minimum execution time. This is due to the nature of Max-Min algoritm (the way how it works) in which it gives more priority to the task with maximum execution time first, leaving tasks with the minimum execution time waiting longer in a queue instead of executing them concurrently. To address these problems, this research proposes three new distributed static batch mode inspired algorithms. The first (proposed) algorithm is based on Min-Min, called Min-Diff, the second algorithm is based on Max-Min, called Max-Average, and the third algorithm is to handle the load balancing, called
Efficient Load Balancing (ELB). In the Min-Diff algorithm, an Initial Task Queue (ITQ) (in non-decreasing order) is generated, where the differences between
maximum and minimum execution time is calculated and compared with the minimum completion time. An appropriate resource for scheduling is selected accordingly. In the Max-Average algorithm, tasks are allocated to resources on the
basis of Average Completion Time (AvCT). In ELB algorithm, the tasks are distributed among resources based on their execution time range. We simulate our
proposed algorithms using a Java based simulator that is purposedly built for Grid computing simulations. The performances of the algorithms are evaluated using several metrics: makespan, average resource utilisation, flow-time, fitness, and load balancing. The results of our proposed algorithms has been compared with the ones
that produced by the standard benchmark algorithms (MCT, MET, Min-Min and
Max-Min). Experimental results demonstrate that the proposed algorithms (Min-
Diff, Max-Average and ELB) are able to produce good quality solution when
compared with the existing algorithm. |
|---|