Hybrid ant colony system algorithm for static and dynamic job scheduling in grid computing
Grid computing is a distributed system with heterogeneous infrastructures. Resource management system (RMS) is one of the most important components which has great influence on the grid computing performance. The main part of RMS is the scheduler algorithm which has the responsibility to map submit...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | eng eng |
| Published: |
2015
|
| Subjects: | |
| Online Access: | https://etd.uum.edu.my/5382/1/s93630.pdf https://etd.uum.edu.my/5382/7/s93630_abstract.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Grid computing is a distributed system with heterogeneous infrastructures. Resource
management system (RMS) is one of the most important components which has great influence on the grid computing performance. The main part of RMS is the scheduler algorithm which has the responsibility to map submitted tasks to available resources. The complexity of scheduling problem is considered as a nondeterministic polynomial complete (NP-complete) problem and therefore, an intelligent algorithm is required to achieve better scheduling solution. One of the prominent intelligent algorithms is ant colony system (ACS) which is implemented widely to solve various types of scheduling problems. However, ACS suffers from stagnation problem in medium and large size grid computing system. ACS is based on exploitation and exploration
mechanisms where the exploitation is sufficient but the exploration has a deficiency. The exploration in ACS is based on a random approach without any strategy. This study proposed four hybrid algorithms between ACS, Genetic Algorithm (GA), and Tabu Search (TS) algorithms to enhance the ACS performance. The algorithms are ACS(GA), ACS+GA, ACS(TS), and ACS+TS. These proposed hybrid algorithms
will enhance ACS in terms of exploration mechanism and solution refinement by
implementing low and high levels hybridization of ACS, GA, and TS algorithms. The proposed algorithms were evaluated against twelve metaheuristic algorithms in static (expected time to compute model) and dynamic (distribution pattern) grid computing
environments. A simulator called ExSim was developed to mimic the static and dynamic nature of the grid computing. Experimental results show that the proposed algorithms outperform ACS in terms of best makespan values. Performance of ACS(GA), ACS+GA, ACS(TS), and ACS+TS are better than ACS by 0.35%, 2.03%, 4.65% and 6.99% respectively for static environment. For dynamic environment,
performance of ACS(GA), ACS+GA, ACS+TS, and ACS(TS) are better than ACS by 0.01%, 0.56%, 1.16%, and 1.26% respectively. The proposed algorithms can be used to schedule tasks in grid computing with better performance in terms of makespan. |
|---|