Dynamic decentralized self organizing scheduling with diana (data intensive and network aware) scheduler

Abstract

Grid Computing due to its effective sharing of heterogeneous resources, economy and portability has become the center of attention for execution of resource hungry divisible processes. But along with the promising characteristics of grid computing it also poses great challenges in its implementation due to the geographical distant resources owned by individuals with differing access and cost policies. The vast applications of Grid Computing in the field of e-sciences gave rise to the need for bulk scheduling (i.e. scheduling of a bulk of jobs as a unique entity). Splitting the bulk may result in a very large number of jobs, making it a hideous job for schedulers and also very time consuming in case of centralized schedulers. In order to exploit the true potential of grid computing workloads need to be scheduled efficiently amongst the participating machines. Centralized schedulers have been implemented to perform the job of load balancing but the grid as a whole lacks the essence of autonomy and self organization. By introducing decentralized schedulers, the limitations such as scalability, posed by centralized schedulers can be tackled. And dependency on a central scheduler is overcome by applying lower level autonomous schedulers. A decentralized approach for bulk scheduling at site level and subgrid level by deploying autonomous site level and low level schedulers is proposed.