Hardware-based, high-throughput data compression is a promising approach to reduce data movement costs in largescale systems and networks, thereby improving overall performance and power efficiency. However, since a data compression algorithm is only effective if it is suitable for the characteristics of the target data, data compression hardware with a fixed algorithm is unrealistic for use in general-purpose environments. To address this challenge, we have been researching adaptive bandwidth compression (ABC) hardware that flexibly provides an effective algorithm depending on the input data. This paper presents the design and the Chisel-based implementation of the ABC hardware platform for encoding compressed data and generating output blocks, as well as introducing a quantization parameter to reduce the circuit area. Our evaluation shows that the proposed quantization parameter can not only reduce the hardware cost, but also control the trade-off between the effective compression ratio and the hardware cost. In addition, based on the evaluation results, we discuss the design optimization of the ABC hardware platform.
各種検索
サポート
ヘルプ