Cloud Computing. 6th International Conference, CloudComp 2015, Daejeon, South Korea, October 28-29, 2015, Proceedings

Research Article

A Buffer Cache Algorithm for Hybrid Memory Architecture in Mobile Devices

Download
380 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-38904-2_30,
        author={Chansoo Oh and Dong Kang and Minho Lee and Young Eom},
        title={A Buffer Cache Algorithm for Hybrid Memory Architecture in Mobile Devices},
        proceedings={Cloud Computing. 6th International Conference, CloudComp 2015, Daejeon, South Korea, October 28-29, 2015, Proceedings},
        proceedings_a={CLOUDCOMP},
        year={2016},
        month={5},
        keywords={Hybrid memory system Buffer cache algorithm Mobile device Foreground application Background application},
        doi={10.1007/978-3-319-38904-2_30}
    }
    
  • Chansoo Oh
    Dong Kang
    Minho Lee
    Young Eom
    Year: 2016
    A Buffer Cache Algorithm for Hybrid Memory Architecture in Mobile Devices
    CLOUDCOMP
    Springer
    DOI: 10.1007/978-3-319-38904-2_30
Chansoo Oh,*, Dong Kang1,*, Minho Lee1,*, Young Eom1,*
  • 1: Sungkyunkwan University
*Contact email: chansoo.oh@skku.edu, kkangsu@skku.edu, kkangsu@skku.edu, yieom@skku.edu

Abstract

In general computing environments including mobile devices, buffer cache algorithm is generally used to mitigate the performance gap between CPU and secondary storage. However, traditional DRAM-based buffer cache architecture reveals a power consumption problem in mobile devices, because it periodically performs the refresh operations to maintain data in DRAM. In addition, traditional buffer cache algorithms never consider the states of mobile applications (e.g., foreground and background state). In this paper, we propose a novel buffer cache algorithm, which efficiently addresses the above issues based on hybrid main memory architecture that is comprised of DRAM and PCM. Our algorithm is motivated by key observation that background applications on mobile device rarely issue I/O requests as well as they can degrade the performance of foreground applications because of the interferences among the I/O requests of applications. For evaluation, we implemented our algorithm and compared its performance against two other algorithms. Our experimental results show that our algorithm reduces the elapsed time of the foreground applications by 53 % on average and the power consumption by 23 % on average without any negative performance effects on background applications.