Overview of Thesis

Distributed collaborative applications are on the immediate horizon for new uses of the World Wide Web. Interaction with data and collaboration over the Web will characterize the next wave of Web applications. To facilitate development of such applications, we propose that functionality common to all such systems be united through shared data spaces. As a data substrate for DCAs, shared data spaces fill the needs of DCAs to share data, allow client-side updates, and ensure consistency of shared data. For good performance and robustness, shared data spaces should employ caches using an eventual consistency protocol that allows unsynchronized access to data. For graphical DCAs with weaker consistency requirements, weakly ordered eventual consistency is easy to implement efficiently.

The highlights of this thesis work are:

• a generic API for handling client updates to shared data consisting of arbitrary fine-grained objects;
• Relay, a multithreaded, hierarchical shared data space that implements weakly ordered eventual consistency under the generic API;
• Multiart, a Relay client application that lets users collaborate graphically with basic shapes using a shared whiteboard model;
• experimentation with Java as a programming language for the Web.

This thesis focuses on weakly ordered eventual consistency as an efficient, reliable, scalable, and acceptable way to propagate updates in a shared data space for a Web-based DCA.

Together, Multiart and Relay illustrate one point in the design space of Web-based DCAs and shared data space substrates. They are meant to form a testbed for different implementations of shared data space properties, from which fully functional shared data spaces could evolve.

This thesis is organized as follows:

• Chapter 2 discusses related work and outlines the evolution of shared data spaces. It starts with a discussion of some distributed systems that exhibit the three traits common to DCAs: shared state, client-side updates, and data consistency. It also explains in more detail why eventual consistency with unsynchronized access is a good choice for a consistency model for shared data spaces for the Web.
• Chapter 3 is a design overview of Relay, a simple shared data space for a Web-based DCAs, focusing on Relay's client/server structure and cache management.
• Chapter 4 details the implementation of the Relay system and Multiart client. This chapter includes a closer look at the use of Java in developing this system. It also details the interface for handling updates in DCAs based on the Relay shared data space.
• Chapter 5 presents conclusions and ideas for future work. We evaluate weakly ordered eventual consistency with unsynchronized access in Web-based DCAs, the use of Java as a programming language for the Web, and the role of shared data spaces as an efficient data substrate for Web-based DCAs.