Adapting a Radial Basis Functions Framework for Large-Scale Computing

This work is aimed at extending a parallel computing framework for radial basis functions methods for solving partial differential equations. Existing framework uses Task Based parallelization method in shared memory architectures to run tasks concurrently on multi-core machines using POSIX Threads. In this method, an algorithm is viewed as a set of tasks each of which performs a specific part of that algorithm while reading some data and producing others. All the dependencies between tasks are translated into data dependencies which makes the tasks decoupled. This work uses the same method but for distributed memory systems using message passing scheme of inter-process conversations. These frameworks cooperates with each other for distributing and running the tasks among nodes and/or cores in a hybrid way of multi-threading and message passing parallel programming paradigms. All the communication between processes (nodes) are performed asynchronously (non-blocking) to be overlapped with computations and the execution flow of the framework is implemented using state machine software construct