Automatic correction of hardware systems based on stream processing

点击翻译按钮获取中文摘要

In big data analysis, such as analyzing data on images, web manipulations, and sensor networks, same operations are repeated on each data item. Such big data are arranged as very long flows of data, called data stream, so that operations on different data items can be processed in parallel as highly pipelined operations. We have developed a new methodology where such data steams are processed by multiple retrievals of relatively small tables. As all tables are retrieved in parallel, we can expect huge parallelisms. It is also called memory-based computing in contrast to processor-based processing which is widely used today. We have developed a set of techniques by which the processing system can be automatically corrected even if some of the tables are disorder by adjusting other tables appropriately.