The Parallel Problems Server

Introduction We describe a novel architecture for a "linear algebra server" that operates in parallel on extremely large matrices. Matrices are created by the server and distributed across many machines. All operations therefore take place automatically in parallel. The server is extenisble and includes a general application interface to clients. This project is motivated by three observations. First, many widely-used algorithms in Computer Science can be realized as operations on matrices. Common techniques in smart text retrieval, object recognition and machine learning can all be described in this framework. Second, it is of utmost importance to be able to test new ideas quickly in an interactive setting. Finally, in order to understand the computational issues that arise with many algorithms it is necessary to test them on very large problems. There are commonly available solutions that address this general problem, but there are several difficulties. Interactive and easi