HIGH-PERFORMANCE INFORMATION PROCESSING IN DISTRIBUTED COMPUTING SYSTEMS

摘要

This paper explores distributed computing systems that may be used efficiently in information processing that is frequently needed in electronic, environmental, medical, and biological applications. Three major components of such systems are: 1) data acquisition and preprocessing; 2) transmitting the results of preprocessing to a higher level computing system that is a PC; and 3) post processing in higher level computing system (in the PC). Preprocessing can be done in highly parallel accelerators that are mapped to reconfigurable hardware. The core of an accelerator is a sorting/searching network that is implemented either in an FPGA or in a programmable system-on-chip (such as Zynq devices). Data is transmitted to a PC through a high-bandwidth PCI-express bus. The paper suggests novel solutions for sorting/searching networks that enable the number of data items that can be handled to be significantly increased compared to the best known alternatives, maintaining a very high processing speed that is either similar to, or higher than in the best known alternatives. Preprocessing can also include supplementary tasks, such as extracting the minimum/maximum sorted subsets, finding the most frequently occurring items, and filtering the data. A higher level computing system executes final operations, such as merging the blocks produced by the sorting networks, implementing higher level algorithms that use the results of preprocessing, statistical manipulation, analysis of existing and acquired sets, data mining. It is shown through numerous experiments that the proposed solutions are very effective and enable a more diverse range of problems to be solved with better performance.