Dynamic deployment of web services on the internet or grid

Abstract

This thesis focuses on the area of dynamic Web Service deployment for grid and Internet applications. It presents a new Dynamic Service Oriented Architecture (DynaSOAr) that enables the deployment of Web Services at run-time in response to consumer requests. The service-oriented approach to grid and Internet computing is centred on two parties: the service provider and the service consumer. This thesis investigates the introduction of mobility into this service-oriented approach allowing for better use of resources and improved quality of service. To this end, it examines the role of the service provider and makes the case for a clear separation of its concerns into two distinct roles: that of a Web Service Provider, whose responsibility is to receive and direct consumer requests and supply service implementations, and a Host Provider, whose role is to deploy services and process consumers' requests on available resources. This separation of concerns breaks the implicit bond between a published Web Service endpoint (network address) and the resource upon which the service is deployed. It also allows the architecture to respond dynamically to changes in service demand and the quality of service requirements. Clearly defined interfaces for each role are presented, which form the infrastructure of DynaSOAr. The approach taken is wholly based on Web Services. The dynamic deployment of service code between separate roles, potentially running in different administrative domains, raises a number of security issues which are addressed. A DynaSOAr service invocation involves three parties: the requesting Consumer, a Web Service Provider and a Host Provider; this tripartite relationship requires a security model that allows the concerns of each party to be enforced for a given invocation. This thesis, therefore, presents a Tripartite Security Model and an architecture that allows the representation, propagation and enforcement of three separate sets of constraints. A prototype implementation of DynaSOAr is used to evaluate the claims made, and the results show that a significant benefit in terms of round-trip execution time for data-intensive applications is achieved. Additional benefits in terms of parallel deployments to satisfy multiple concurrent requests are also shown.