For the near future, we are working on adding a few more useful features to the current framework. As mentioned in the previous section, a caching or pre-fetching engine that minimizes communication or overlaps communication with computation would be very useful for improving overall performance. We are also working on adding a request feature to the watcher client that would allow users to ask the server to use different target parameters. For example, for the web crawling application, we can add a text box to the watch GUI through which the user can specify a new starting web page to the server. This may be implemented by adding a request() method in the problem object, and having the watch manager call it in response to a request from the watcher client.
While it is easy to create different master-worker style applications in the current implementation, it is not yet easy to implement other programming models because of some unnecessary couplings between lower-level components. For example, currently, the BasicWorkerApplet must specify the type of work engine to create, so using a new type of engine requires writing a new client applet class. If we can instead have the applet load the appropriate type of engine at runtime, we can just have single applet class for all applications, and the type of engine it uses is specified to it by the server according to the problem. We currently have a new implementation which supports this and other flexibility features. However, we still need to reimplement some basic components, and so have not yet been able to get any experimental results.
Once the new implementation of the framework is completed, we intend to experiment not only with new programming models such as Cilk, but also with security mechanisms such as checksums.