A Policy-Based Resource Brokering Environment for Computational Grids

With the advances in networking infrastructure in general, and the Internet in particular, we can build grid environments that allow users to utilize a diverse set of distributed and heterogeneous resources. Since the focus of such environments is the efficient usage of the underlying resources, a critical component is the resource brokering environment that mediates the discovery, access and usage of these resources. With the consumer\u27s constraints, provider\u27s rules, distributed heterogeneous resources and the large number of scheduling choices, the resource brokering environment needs to decide where to place the user\u27s jobs and when to start their execution in a way that yields the best performance for the user and the best utilization for the resource provider. As brokering and scheduling are very complicated tasks, most current resource brokering environments are either specific to a particular grid environment or have limited features. This makes them unsuitable for large applications with heterogeneous requirements. In addition, most of these resource brokering environments lack flexibility. Policies at the resource-, application-, and system-levels cannot be specified and enforced to provide commitment to the guaranteed level of allocation that can help in attracting grid users and contribute to establishing credibility for existing grid environments. In this thesis, we propose and prototype a flexible and extensible Policy-based Resource Brokering Environment (PROBE) that can be utilized by various grid systems. In designing PROBE, we follow a policy-based approach that provides PROBE with the intelligence to not only match the user\u27s request with the right set of resources, but also to assure the guaranteed level of the allocation. PROBE looks at the task allocation as a Service Level Agreement (SLA) that needs to be enforced between the resource provider and the resource consumer. The policy-based framework is useful in a typical grid environment where resources, most of the time, are not dedicated. In implementing PROBE, we have utilized a layered architecture and façade design patterns. These along with the well-defined API, make the framework independent of any architecture and allow for the incorporation of different types of scheduling algorithms, applications and platform adaptors as the underlying environment requires. We have utilized XML as a base for all the specification needs. This provides a flexible mechanism to specify the heterogeneous resources and user\u27s requests along with their allocation constraints. We have developed XML-based specifications by which high-level internal structures of resources, jobs and policies can be specified. This provides interoperability in which a grid system can utilize PROBE to discover and use resources controlled by other grid systems. We have implemented a prototype of PROBE to demonstrate its feasibility. We also describe a test bed environment and the evaluation experiments that we have conducted to demonstrate the usefulness and effectiveness of our approach