Intelligent SSD: A turbo for big data mining

Duck-Ho Bae, Jin-Hyung Kim, Yong-Yeon Jo, Sang-Wook Kim, Hyunok Oh, Chanik Park

This paper introduces a new notion of the intelligent SSD and presents its potential benefits in terms of data mining applications. With intelligent SSDs, a large volume of data can be directly processed by CPU and DRAM inside intelligent SSDs, and the final result of a very small size needs to be transferred to the host CPU instead of all the data stored in intelligent SSDs. We first discuss design considerations of intelligent SSDs compared with the current SSD architecture. We then analyze the execution costs of data mining applications running on intelligent SSDs by formulating their cost models. Finally, we show the efficiency of performing data mining on intelligent SSDs by comparing it with those of traditional ones through a series of simulations. Through the experimental results, we show that the intelligent SSDs provide significant improvement in performance over the host CPUs in processing data-intensive data mining applications.