A Performance Based QoS Aware Resource Brokering Framework for the Grid

Most of the current day grid infrastructures operate on a best effort mode, in offering services to the end users. However to cater to the ever increasing performance demands of grand challenge applications, it is imperative for the grid computing technologies to evolve and mature, to deliver services with highest QoS attached with them. The work described in this paper focuses on providing a QoS aware resource brokering framework that can deliver optimal performances for the applications on the grid. Similar to the popular Amazon EC2 instance types of classifying resources based on the application requirements, the framework identifies different profiles of HPC applications, and thus chooses the right mix of QoS metrics to deliver the best possible performance with the available resources of the grid, by making dynamic Service Level Agreements with the resource and service providers. The paper gives an insight into the various performance metrics employed to optimally allocate the resources. Further it includes a detailed discussion of the test results obtained using this framework on a production grid infrastructure, highlighting the perceived performance gains by grid applications.