Using data mining and machine learning techniques for system design space exploration and automatized optimization
Yuxuan Wang, Qihui Sun, Ting‐Ying Chien, Po-Chun Jimmy Huang · 2017
Recently, the significance of data mining and machine learning have been highlighted in diversified application scenarios. Various data mining and machine learning techniques are often used to analyze the gigantic amount of data to create more commercial values in high-end enterprise systems. However, the advancement of technologies has made data mining and machine learning possible on low-end systems, such as personal computers or embedded systems. While researchers have proposed excellent work on the management de-signs of different components of the system, most of the work are built upon the characteristics of the system, which may change from time to time. This makes it impossible to optimize the system performance with stat-ic, or statically adaptive, system designs. In this work, we propose to embed the supports of data mining and machine learning to the design of operating system, so as to discover a new, automatized way to adaptively optimize the system without using complex algorithms. To validate the proposed ideas, we choose the cache design as a case study, where the replacement of cached contents is automatically controlled by a decision maker. The decision maker then replies on a data miner, which analyzes the data collected by the system monitor. The efficacy of the considered case is verified by a series of experiments, where the results are quite encouraging.