Energy-efficient response time management for embedded databases
详细信息 查看全文
- 作者：Woochul Kang ; Jaeyong Chung
- 关键词：Embedded databases ; DVFS ; Quality ; of ; data ; Feedback control ; Quality ; of ; service
- 刊名：Real-Time Systems
- 出版年：2017
- 出版时间：March 2017
- 年：2017
- 卷：53
- 期：2
- 页码：228-253
- 全文大小：
- 刊物类别：Computer Science
- 刊物主题：Computer Systems Organization and Communication Networks; Communications Engineering, Networks; Special Purpose and Application-Based Systems; Performance and Reliability; Control, Robotics, Mechatron
- 出版者：Springer US
- ISSN：1573-1383
- 卷排序：53
文摘
Recently, as the amount of data from both the cyber and physical worlds are exploding, database functionalities are increasingly embedded into mobile/embedded systems to provide systematic in situ data services. Such databases, often called embedded databases, are supposed to provide timely data services under various constraints, such as limited time and energy. However, current embedded databases do not support timely and energy-efficient processing of transactions. Further, most existing academic results from real-time and embedded databases depend on the adaptation of quality-of-data (QoD) and admission control techniques to achieve timeliness. However, the range of QoD adaptation is often limited by the requirements of applications. We might consider different tuning knobs exposed by modern embedded systems, such as dynamic voltage/frequency scaling (DVFS), since they can be applied to control the latency of transactions without degrading QoD as far as the system is not overloaded. However, since DVFS has only a few discrete voltage/frequency steps, fine-grained control to support desired response time is hard to achieve. In this paper, we present a method of combining the advantages of these two very different control knobs to achieve the energy-efficiency while still supporting fine-grained control of response time of transactions. We formally design the feedback control loop using these two complementing control knobs, and demonstrate the feasibility of our approach by implementing it in a modern embedded device. The experimental results show that our approach closely supports the desired response time of transactions while achieving lower energy consumption compared to baseline approaches.