Scalable and efficient resource management with proximity-awareness for computational grids

The Internet provides high speed access to computing resources that are geographically dispersed and are heterogeneous in nature. This pool of resources requires management techniques that will propagate the continued success of Internet entities, such as Grid and P2P applications. While managing these geographically dispersed resources via queries, attention must be given to avoid the possibility of hotspots that could develop in areas containing objects holding resource data. One approach to avoid the possibility of hotspots is to replicate resource data and give querying nodes the most efficient choice possible for query resolution during the querying process. This approach should locate resource data as close as possible to potential resource requesting nodes. Therefore, data replication is a desirable trait for a resource management system.;In this dissertation, I will consider a two-layered approach to managing these resources. The two layers are differentiated by the roles of a (1) resource management scheme, and (2) data replication scheme. The resource management scheme is called FaSReD, which is a Fast and Scalable Resource Discovery System in Support of Multiple Resource Range Requirements for Computational Grids. FaSReD provides a network traffic management technique that only requires a limited storage space for a given resource contribution. FaSReD accomplishes this feat by bit string encoding a resource and placing this bit string on the network and relying on range queries to locate these desired resources.;The data replication scheme that accommodates and enhances FaSReD is called PADME, which is an effective P2P replica management system for distributed hash tables based on proximity-aware distributed mutual exclusion. PADME gives consideration to the location of replicated resource management data in an effort to curb ongoing network traffic. PADME will enhance the resource discovery process by determining the extent and location of replicated resource data. This enhancement will lead to efficient scheduling of jobs by conveniently providing schedulers with multiple efficient choices. The main feature of these resource scheduling choices will be the proximity of resource data to requesting nodes.