Mahidol University Logo
Faculty of ICT, Mahidol University
 

Admissions

Printable Version

MINIMIZING DATA COMMUNICATION COST IN GRID COMPUTING SYSTEM

 

TITLE MINIMIZING DATA COMMUNICATION COST IN GRID COMPUTING SYSTEM
AUTHOR EKASIT KIJSIPONGSE
DEGREE DOCTOR OF PHILOSOPHY PROGRAMME IN COMPUTER SCIENCE(INTERNATIONAL PROGRAMME)
FACULTY FACULTY OF INFORMATION AND COMMUNICATION TECHNOLOGY
ADVISOR SUDSANGUAN NGAMSURIYAROJ
CO-ADVISOR DAMRAS WONGSAWANG
VASAKA VISOOTTIVISETH
 
ABSTRACT
Grid computing is a paradigm of distributed computing that consolidates remote and heterogeneous computing resources to build a large virtual computing environment. Users can access the combined capabilities of diverse Grid computing resources as if they were a single computer. The advent of Grid computing attracts scientists and researchers to develop new classes of applications which demand intensive computation to analyze a large amount of data located across a distributed computing environment. However, due to low bandwidth and high latency in a wide-area network, the perfor- mance of such applications could be dominated by the data communication cost rather than the computation cost. This thesis investigates the key issues to improve the performance of Grid applications via the minimization of the communication cost in tandem with the compu- tation cost. Even though there are many relevant issues and nontrivial problems, three challenging problems are addressed in this work. The first challenge is to develop a data partitioning algorithm for tightly-coupled parallel applications in which each node performs a computation on a subset of data while communicating with its neighboring nodes for transferring intermediate outputs. We propose a heuristic algorithm to parti tion high dimensional data so that the aggregated communication and computation cost is reduced. The second challenge deals with the balance between communication and computation costs when placing pipeline applications over distributed Grid resources. We propose two algorithms that assign pipeline stages to nodes in order to minimize the computation cost and to prevent the degradation of the pipeline throughput caused by the communication cost between nodes. The last challenge focuses on the problem of data transfer scheduling in a Grid environment that has multiple replicated sources and transfers data through multiple paths. We propose an efficient distributed scheduling framework to coordinate network resource allocation for multiple data transfers to reduce the average completion time. All of the proposed algorithms have been evaluated and the results are convincing for use in practice. In conclusion, the main contributions of this thesis are new and efficient approaches to manage the communication cost to improve the performance of Grid applications.
KEYWORD GRID COMPUTING / PIPELINE PLACEMENT / DATA PARTITIONING / TRANSFER SCHEDULING

 

Go to Top

 

ICT Building, Mahidol University, 999 Phuttamonthon 4 Road, Salaya, Nakhonpathom 73170 Tel. +66 02 441-0909 Fax. +66 02 849-6099
Mahidol University Computing Center, The Faculty of ICT, Mahidol University , Rama 6 Road, Rajathevi, Bangkok 10400 Tel. +66 02 354-4333 Fax. +66 02 354-7333