Many software companies are migrating to server based component systems. Such distributed systems use thin clients. This gives lower administration costs, permits updates to be performed automatically and gives a facilitated deployment.
One of the companies that are adapting to their software to this way of thinking is Ida Systems Ab in Linköping. Most of Ida's products are built on their own platform PAX Enterprise, which now is to be transformed into a server based component system. However Ida wants to avoid the biggest drawback with thin clients, the waiting times experienced by users as components are downloaded from the server. Such perceived waiting times cause irritation and inefficiency and must therefore be minimised.
This thesis presents a solution to waiting time problem in thin clients, by using intelligence at the server to model and predict users' behaviour. Predictions are used to push components to a client cache before any explicit requests have been made. If a prediction is correct and the corresponding push operation finishes before the components are needed, the user will have the feeling thatshe uses a local program. In simulations based on PAX Enterprise usage it has been shown in this thesis that the proposed distribution model can cut down the waiting times by nearly 60%.
