A multimedia parallel processing approach on GPU MapReduce framework

At present, most of the GPU MapReduce frameworks are based on single multimedia processing program, and cannot be used to handle multiple multimedia processing programs simultaneously. The service needs for multiple multimedia processing programs can only be satisfied by sequencing, which lacks efficient data segmentation and resource scheduling management. As a result, the hardware efficiency is reduced under the multiple multimedia processing programs. Based on the existing MapReduce framework of GPU, Mars, this study designed a parallel processing mechanism for multiple multimedia processing programs. According to the processing needs of current multimedia processing programs, hardware resources demand, and data processing capacity, the proposed mechanism segments the large quantity of data produced by multiple multimedia processing programs, and transmits the suitable work load segments according to the hardware loading capacity for further processing. This study uses the multimedia processing program, which is commonly used for MapReduce framework computation, as the experimental work load, and treats the execution time as the index for the efficiency improvement. The results suggest that the average processing speed under the proposed mechanism is improved by 1.3 times.