PALATIN: A Platform for Interactive Algorithms


Writing distributed programs is a challenging task. Many researchers have tried to overcome this problem with the help of a suitable programming environment, using message passing in the 1970s, remote procedure calls in the 1980s, and distributed shared memory (DSM) in the 1990s. Still, the percentage of distributed applications based on these environment is very small. One possible xplanation for this is that each round is an evolutionary stage for the distributed programming paradigm, and each attempt brings us closer to a methodology that will ultimately be easy to use and efficient. A less optimistic explanation is that no matter what methodology people will come up with, ease of use and efficiency are two concepts that cannot be achieved at the same time. We believe that this is in fact true for the direction in which the design of universal platforms for distributed computing is currently heading. The basic dilemma is that procedural thinking and the von Neumann machine have dominated our way of designing and thinking about programs for distributed environments, but these approaches are inadequate for distributed computing. Thus, a different paradigm is needed, which we call the Spheres model (which is similar in flavor but much more powerful than cellular computing models). To illustrate the basic ideas behind Spheres, we constrast it with DSM models. The DSM approach suffers from the following deficiencies: The deficits above reveal that major changes should be done in order to arrive at a paradigm useful for distributed applications: These are the basic ingredients of the Spheres model. The goal of this project is to implement a PALATIN platform that allows the easy design of efficient programs under this Spheres model. An early Spheres simulation environment has already been used in the course "Theory of Network Communication" in the fall 2003 semester, and an improved environment will be used for the same course in the fall 2004 semester.

Faculty members:

PhD students:



Christian Scheideler
Last modified: Wed June 2 2004