Exploiting Reconfigurable FPGA for Parallel Query Processing in Computation Intensive Data Mining Applications

This work concentrates on exploiting re-configurable Field Programmable Gate Arrays (FPGAs), an SRAM-based FPGA coprocessor, for query processing in computation-intensive data mining appli cations. Complex computation-intensive data mining applications in geoscientific and medical information systems environments often require support for extensibility and parallel processing to deli ver the necessary functionality and hi gh performance. Emerging FPGA technology represents a promising hybrid hardware/software (HW/SW) co-design approach [20,21] to augment traditional query processing techniques by efficient use of reconfigurable task-specific hardware kernels and host processor(s). In this work, we study the properties and characteristics of a FPGA co-processor system, VCCEVC, which uses a Xilinx 4020E part and a slaveinterface for data acquisition. We present a simple HW/SW cost model for parallel query processing. And also, we study a HW/SW partitioning problem by looking into the NASA Cyclone-Tracking Data Mining appli cation. We identify the most computation intensive routines in this application that runs on our extensible parallel geoscientific query-processing environment, Conquest. Finall y, we discuss the applicability of the hybrid approach for the given application based on our results.