OpenCL Implementation of Gzip on Field-Programmable Gate-Arrays

Presented by Mohamed S. Abdelfattah, University of Toronto and Altera

[gview file=”/wp-content/uploads/iwocl-2014-tech-presentation-Mohamed-Abdelfattah.pdf”]
Hardware implementation of lossless data compression is important for
optimizing the capacity of storage devices in data centers, and bandwidth and
power of communication channels in high-speed networks. In this work we use the
Open Computing Language (OpenCL) to implement the two parts of Gzip compression
(LZ77 compression and Huffman encoding) on a field-programmable gate-array
(FPGA). We show how we make use of a heavily-pipelined custom hardware
implementation of our compression kernels to achieve very high compression
throughput — 3 GB/s for LZ77 kernel that uses ~50% of a Stratix-V A7 chip. We
achieve an average compression ratio of 1.7X with LZ77 compression alone, and
2.1X when we include a simple version of Huffman encoding.