Coallocation of compute and storage resources in grid environments

This thesis considers the problem of availability of storage space for job executions in a grid environment. A number of applications in science and engineering require large amounts of input data and hence, storage space, for computations. The computational and storage needs of such applications in a grid environment are satisfied by the compute facilities, shared among multiple users. As compute facilities have a fixed amount of storage space, are shared among several users and serve applications with large input data, the storage space often becomes bottleneck. The storage space can also become bottleneck due to failure or improper removal of data after job executions. Job execution requests fail when storage space is unavailable because the required input data cannot be transferred to the compute site. The storage space needs to be properly allocated to the jobs to deal with the unavailability.; A co-allocation system is developed that enables simultaneous and overlapped allocation of storage space to the jobs along with the compute processors. Strategies and schemes are developed that automate job executions by dynamically allocating storage space, performing data transfers before job executions and de-allocating space after executions. A lease based storage manager, required by the co-allocation system, is also developed that actually manages, tracks availability, allocates and de-allocates the storage space.; The co-allocation schemes are designed to allocate storage space at the earliest suitable time, minimize storage space occupancy durations and de-allocate the storage space soon after job executions. The schemes also minimize the job turnaround times by transferring the input data while jobs wait for availability of processors and vice-versa.