Contributions to distributed objects and network agents
Abstract
This dissertation presents the theoretical foundations and implementation of the Bond distributed object system and agent framework. In Bond we put the emphasis on dynamic composition and runtime configuration. Every object has communication capabilities and can become both a client and a server. Objects can be extended during runtime to provide new functionality with probes. Security, monitoring and logging capabilities can be configured dynamically. Services can be created upon receipt of a request. Two objects can discover their interfaces dynamically, without resorting to an external service like an interface repository. The component based approach also offers advantages at the agent level. Bond agents can be assembled locally or at remote locations using the blueprint agent description language. The agents can be structurally modified during runtime using surgical blueprint scripts. Agents are mobile (can be moved from one host to another during runtime) and checkpoint-able (their state can be saved in persistent storage and they can be restarted at an arbitrary later time, at the same or different location). A theoretical model for agents is introduced. Using this formalism, agents can evaluate predicates like the satisfiability of the agenda, migratability or reachable subsets of the model. The theoretical model also provides a formal method of evaluating how the surgical operations affect the behavior of agents. This formalism offers the possibility of determining guaranteed invariants for the behavior of agents for certain conditions. Performance measurements were conducted to evaluate the cost of the additional functionality in Bond objects.
Degree
Ph.D.
Advisors
Marinescu, Purdue University.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.