DESIGN AND EVALUATION OF RESOURCE ALLOCATION AND JOB SCHEDULING ALGORITHMS ON COMPUTATIONAL GRIDS

Grid, an infrastructure for resource sharing, currently has shown its importance in
many scientific applications requiring tremendously high computational power. Grid
computing enables sharing, selection and aggregation of resources for solving
complex and large-scale scientific problems. Grids computing, whose resources are
distributed, heterogeneous and dynamic in nature, introduces a number of fascinating
issues in resource management. Grid scheduling is the key issue in grid environment
in which its system must meet the functional requirements of heterogeneous domains,
which are sometimes conflicting in nature also, like user, application, and network.
Moreover, the system must satisfy non-functional requirements like reliability,
efficiency, performance, effective resource utilization, and scalability. Thus, overall
aim of this research is to introduce new grid scheduling algorithms for resource
allocation as well as for job scheduling for enabling a highly efficient and effective
utilization of the resources in executing various applications.
The four prime aspects of this work are: firstly, a model of the grid scheduling
problem for dynamic grid computing environment; secondly, development of a new
web based simulator (SyedWSim), enabling the grid users to conduct a statistical
analysis of grid workload traces and provides a realistic basis for experimentation in
resource allocation and job scheduling algorithms on a grid; thirdly, proposal of a new
grid resource allocation method of optimal computational cost using synthetic and
real workload traces with respect to other allocation methods; and finally, proposal of
some new job scheduling algorithms of optimal performance considering parameters
like waiting time, turnaround time, response time, bounded slowdown, completion
time and stretch time. The issue is not only to develop new algorithms, but also to
evaluate them on an experimental computational grid, using synthetic and real
workload traces, along with the other existing job scheduling algorithms.
Experimental evaluation confirmed that the proposed grid scheduling algorithms
possess a high degree of optimality in performance, efficiency and scalability.