Service Architecture for Heterogeneous Access, Resources, and Applications*

Pervasive computing demands all-encompassing exploitation of services inside the network. Our overarching goal is to understand how to create end-to-end services with desirable and predictable properties, such as performance and reliability, when provisioned from multiple and independent service providers. Services are the components of distributed applications and the glue that interconnects them as they function across the network. These range from providing basic network reachability to creating overlay networks with enhanced qualities like predictable latencies and sustained bandwidths. They also include instances of application building blocks, requiring processing and storage, judiciously placed in the network to control connection latencies and to achieve scale through load sharing. Such services may be simple format translators, interworking functions, or major subsystems for content distribution or Internet search, or demand-response pricing in electricity markets. Composition via interconnection of services allows more sophisticated services and applications to be constructed hierarchically from more primitive ones. Since economics makes it unlikely that any single service provider will be able to provide all of the connectivity, applications building blocks, processing, and storage resources to effectively deploy a globe-spanning application, the composition of services across independent providers is essential. We have been developing a comprehensive framework for introducing new qualities into the Internet's routing framework. These qualities are well beyond traditional quality of service for network flows, and focus instead on the management of the routing infrastructure itself, such as the agility of the network to respond to routing changes due to network component failure or the detection of misconfigured or malicious routers within the network. Specifically, we have: (1) developed new methods for achieving more rapid convergence in response to routing changes, (2) proposed a new "policy plane" for managing the peering relationships between regions to achieve better load balancing and fast failover among alternative network paths, (3) new techniques for trading bandwidth for loss rate to achieve certain kinds of guaranteed services without requiring any support from the network, and (4) an approach for verifying the route advertisements propagating through the network to uncover misconfigured or malicious routers that cause parts of the network address space to become hijacked. These developments are leading to prototypes that we intend to deploy in the PlanetLab testbed, a wide-area experimental overlay network being jointly developed between a number of universities and industrial research laboratories. *This project is not officially supported through CITRIS funds, but the faculty and topical affiliations are sufficiently strong that it is listed here for referral and convenience.