| In this dissertation, we proposed (1) a methodology for scalable and effective control plane in support of QoS communication services over computer networks, and (2) a light weight resource discovery for QoS sensitive software components in distributed systems.; Provisioning QoS guaranteed communication services over large-scale computer networks involves many activities, including QoS routing, resource reservation, admission control, and associated signaling and message exchanges. Our methodology makes the control plane scalable and effective by efficiently managing network resources based on the sink tree paradigm. As shown in the simulation results, the sink tree paradigm provides (1) low signaling overhead, (2) high resource utilization, and (3) fast flow set-up inherently. Also, by making use of an existing link state routing protocol, the extended OSPF, it accommodates network dynamics, such as traffic congestion and topology change by re-configuring the tree and re-allocating network resources. Studies on the sink tree dynamics, based on implementation and measurements, are left for future work.; QoS sensitive software components require application survivability and information assurance under external attack in vulnerable distributed environment. A promising way is migration of the components to another place where either enough resources are available for the migrating components or they can run in higher security levels. Resource discovery therefore plays a central role in such s distributed systems. Our proposal, REALTOR, discovers resources with (1) low protocol overhead, and (2) high information accuracy, based on the dynamic community concept. Each host is free to build its own community for future migration, and at the same time, it is a member of many communities for resource pledging. Simulation results show that REALTOR, an example of pull-based approach, (1) works far better than the traditional push-based approach in terms of protocol overhead, and (2) performs much better in locating resources that the traditional push-based approaches. Currently, we are implementing and measuring performance of REALTOR in Java. Improvement of REALTOR based on implementation results are left for future work. |
