On the design of a demand -based network -computing system: The Purdue University Network-Computing Hubs

Demand-based network-computing systems are characterized by their universal accessibility and their ability to harness networked resources as and when necessary. This dissertation addresses key issues involved in the design of such a system: (1) the interface to the external world, (2) the internal system design, (3) support for legacy applications, and (4) resource management across administrative domains. An operational prototype, the Purdue University Network Computing Hubs (PUNCH), is presented. The interface to the external world is based on a unique network desktop that provides Web access to computing services by treating URLs as locations in a dynamic, virtual, and side-effect based address-apace. The internal system design accounts for scalability and reliability, works across administrative domains, and employs a novel encoding mechanism that allows O(1) access to all managed information. The internal architecture is designed around a three-level hierarchy with replicatable components, a choice driven by the nature of the information associated with run-time tradeoff decisions. Unmodified (legacy) applications are supported by way of a highly flexible tool-specification language and a metaprogramming environment specifically designed to adapt to the needs of different tools (to date, forty tools developed by four vendors and eight universities have been installed on PUNCH). Resource management across administrative domains is enabled by a metaprogramming environment that allows administrators to specify usage constraints and policies for users and resources; cost/performance decisions are made by the network-computing system at run-time on the basis of predicted resource-usage characteristics and specified usage constraints and policies