Optimizing Data Mining Workloads using Hardware Accelerators

Data mining is the process of finding useful and actionable patterns in large data sets. Data mining algorithms have become vital to researchers in science, engineering, medicine, business, search and security domains. In recent years, there has been a tremendous increase in the size of the data being collected and analyzed. Data mining algorithms have been unable to scale up to these vast amounts of data, leading to significant performance degradation. Also, the enhancements in processor and system designs do not necessarily aid data mining workloads. In our previous work, we demonstrated that computational characteristics as well as data access requirements for data mining workloads are quite different than those of other common workloads. Therefore, there is a need to specifically address the limitations of accelerating data mining workloads. In this paper, we present a brief overview of the major challenges faced in data mining systems design. We first highlight important characteristics of these workloads. Then, we describe some initial designs and results for accelerating data mining algorithms using programmable hardware. Our results show that tremendous performance gains can be obtained by accelerating these workloads when compared to using traditional systems.