MoBiDiCK: a tool for distributed computing on the Internet

We have developed a software tool called MoBiDiCK (Modular Big Distributed Computing Kernel) that is ultimately intended for distributed computing. In this paper, we detail the design and show results using the core components of MoBiDiCK running two different clients on a local cluster. MoBiDiCK is a database-driven system that can be used to marshal a large number of processors across the Internet in order to have them collaborate on a single computation. These utilize a message-passing API and control synchronization formalism we have developed that uses the HTTP standard and Web servers. CGI programs on the volunteer processors perform the computations. The problem domains best served by MoBiDiCK are parallel computing problems that are CPU-bound (not I/O-bound) and require minimal inter-process communication. The parallel tasks that we present include the analysis of databases of 3D protein structures and Monte Carlo simulations for ab-initio protein folding.