NEW APPROACH FOR SPLITTING A CLASS OF REAL-TIME ALGORITHMS FOR PIPELINED PROCESSING

This paper aims to develop a generalized technique for splitting a giv en sequential algorithm into a number of sequential stages for pipelin ed mode of execution. Each stage so identified is executed by one proc essor, and the time of execution of each stage is equal to the data ar rival time. If such splitting is done at logical interfaces, a process or needs to communicate to its successor processor partial results onl y. However, such splitting cannot always be carried out at logical int erfaces in the interest of maximizing the throughput. This necessitate s additionally the communication of processor status to the succeeding processor. This problem has been carefully in the proposed scheme and the effectiveness of the scheme has been tested experimentally using a number of real-time algorithms.