Distributed Parallel Computing Using Navigational Programming: Orchestrating Computations Around Data

摘要

Message Passing (MP) and Distributed Shared Memory (DSM) are the two most common approaches to programming on distributed memory systems. MP is difficult to use, while DSM is not scalable. Performance scalability and ease of programming can be achieved at the same time by using "shared variable programming" and following the principle of "computation locus following data," which is embodied in our Navigational Programming approach. The implementation of a real-world algorithm, parallel Cholesky factorization, presented in this paper supports our claim that Navigational Programming is better suited for general purpose distributed programming than either MP or DSM.