Cognitive Science Based Scheduling In Grid Environment

Abstract

Decisions are made in grid scheduling based on complexity of the requirement of the job, either as data or computation intensive. However the major challenges are the data locality, transfer and execution of data in suitable locations. The existing solutions had solved the issues considering only the individual constraints but solutions considering all the challenges are not available, thus demanding the need for a unique solution. A novel solution of introducing cognitive science into grid workflow environment is proposed to reduce the make span by considering all the challenges together. Cognitive Artificial Intelligence is used to develop a machine with intelligence which receives the request for data sets and reduce the size of data sets by partitioning and a unique algorithm, namely Cognitive Mode Algorithm (CMA) is also proposed for effective allocation of data sets based on the request from the user. Here the data sets are partitioned by considering the size and available network bandwidth. The replication of partitioned data sets across different sites is done and stored in partitioned metadata repository. This will minimize the recurrence of partition for the same data request in future. A Data Location Matrix (DLM) is also constructed by having the distance of the site and the details of the data sets stored in that site. Prediction of the next request from the same site is also made to reduce the data availability time. Cognitive mode algorithm focuses on learning, thinking, and perception to show the intelligence. The results shows that it reduces the data transfer time, execution time and data availability time which in turn will reduce the overall make span. © Springer International Publishing Switzerland 2015.