Evaluating Master-Slave Implementation of Double-DOT Parsing for Massive ParallelMIMD Systems

摘要

The performance of the master-slave implementation on a MIMD system of a parallelrecognizer, which uses the double dot algorithm, is evaluated. An overview of general parallel programming techniques for distributed memory MIMD systems is given. The double dot parsing algorithm is explained. The master-slave implementation is discussed and analyzed. Results of the analysis are used and improvements for the implementation, covering both the software and the hardware, are proposed. Conclusions on the performance of the double dot parsing algorithm are given.