RAM-based mergers for data sort and frequent item computation

Data sorting and frequent item computation are important tasks in data processing. The paper suggests an architecture for parallel data sorting with simultaneous counting of every item frequency. The architecture is designed for streaming data and incorporates data sorting in hardware, merging of preliminary sorted blocks with compressing of repeated items with calculating of repetitions in hardware, and merging large subsets received from the hardware in general-purpose software. Hardware merge components of this architecture count and compress repeated items in sorted subsets in order to reduce merging time and prepare the data for frequent item computation. The results of experiments clearly demonstrate advantages of the proposed architectures.