Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

The analysis and synthesis of a parallel sorting engine

Thesis/Dissertation ·
OSTI ID:6093165
This thesis is concerned with the development of a unique parallel sort-merge system suitable for implementation in VLSI. Two new sorting subsystems, a high performance VLSI sorter and a four-way merger, were also realized during the development process. In addition, the analysis of several existing parallel sorting architectures and algorithms was carried out. Algorithmic time complexity, VLSI processor performance, and chip area requirements for the existing sorting systems were evaluated. The rebound sorting algorithm was determined to be the most efficient among those considered. The rebound sorter algorithm was implemented in hardware as a systolic array with external expansion capability. The second phase of the research involved analyzing several parallel merge algorithms and their buffer management schemes. The dominant considerations for this phase of the research were the achievement of minimum VLSI chip area, design complexity, and logic delay. It was determined that the proposed merger architecture could be implemented in several ways. Selecting the appropriate microarchitecture for the merge, given the constraints of chip area and performance, was the major problem. The tradeoffs associated with this process are outlined. Finally, a pipelined sort-merge system was implemented in VLSI by combining a rebound sorter and a four-way merger on a single chip. The final chip size was 416 mils by 432 mils. Two micron CMOS technology was utilized in this chip realization. An overall throughput rate of 10M bytes/sec was achieved.
Research Organization:
Oregon State Univ., Corvallis, OR (USA)
OSTI ID:
6093165
Country of Publication:
United States
Language:
English