Advance Reservation-based Computational Resource Brokering using Earliest Start Time Estimation

As a potential method for improving the resource utility, advance reservations is crucial for brokering computational resource in a predictable and efficient way. However, Conducting Advance Reservation efficiently in grid environment is not easy due to the insufficient support of the current grid architecture and the dynamic feature of grid resources. Based on a grid middleware Migol, this paper presents and evaluates a job broker architecture that addresses this deficit and provides support for advance reservation on grid-level. Using Earliest Start Time Estimation algorithm (ESE) to predict queuing delays at grid resources, this architecture is able to conduct advance reservation at lower cost and higher accuracy. Experiments show that this brokering architecture is scalable and the ESE algorithm is efficient in use.