Parallel Computation of Linear Recursive Queries in Knowledge-Based Systems

Parallel processing of recursive queries in KB systems is an important means to improve performance. However, two significant problems: heavy interprocessor communication and poor load balancing, are inherent features of query answering methods based on Least Fixed Point computation, such as Naive and Semi-Naive. They make it difficult to ensure efficiency for all distributions of data values. Alternative parallel methods are needed, whose performance is less dependent on the data. This paper examines parallel aspects of the Arithmetic Method, which offers a solution to the communication and load balancing problems. A prototype has been built. The experimental result is analysed. The advantages are: i) no interprocessor communication during task processing, ii) automatic data partitioning, iii) easy load-balancing, iv) high utilisation of processors.