A parallel numerical algorithm for boundary-value FIDEs on a PC cluster

An algorithm for parallel processing the discrete nonlinear system for solving Fredholm integro-differential equations with two-point boundary conditions on a PC cluster is considered. The cost of calculating the history terms is expensive and improvements are motivated by considering different architectures. The algorithm has been modified to minimize the communication overhead inherent in a distributed application and tested using two different messaging protocols: GAMMA, an efficient message system for clusters of PCs (Ciaccio, 1999) with low latency and high throughput and TCP/IP protocol with MPI. Numerical examples illustrate the results.