A bit-level matrix transpose for bitmap-index-based data analytics

When it comes to data analytics, bitmap index is likely to be one of the most effective approaches to processing a large number of queries with multiple conditions. In this work, we propose a ping-pong architecture of a bit-level matrix transpose (BLMT), which is successfully applied to the creation of bitmap index. BLMT employs a set of registers and multiplexers operating in parallel, so as to transpose every single row of matrix within one clock cycle. The experimental results in an Altera Arria V SX FPGA show that BLMT achieves the peak throughput of 6.37 Gbps at 50-MHz operating frequency, which is equivalent to 99.5% of the theoretical throughput. Additionally, BLMT only costs approximately 1.5% of lookup tables and 4.8% of registers.