• DocumentCode
    2106974
  • Title

    A Light-Weight Scheduler for Single-ISA Asymmetric Multi-core Processor Using Online Profiling

  • Author

    Xu, Yuanchao ; Du, Lin ; Zhang, Zhimin

  • Author_Institution
    Inst. of Comput. Technol., Beijing, China
  • fYear
    2011
  • fDate
    6-8 July 2011
  • Firstpage
    183
  • Lastpage
    188
  • Abstract
    Single-ISA asymmetric multi-core processors are thought as next generation emerging multi-core architecture which has better performance per watt over symmetric multi-core processors. However, they also pose unique challenges to operating system, which traditionally assume homogeneous hardware. These challenges include how to analyze program behavior dynamically in kernel with low overhead, how to modify traditional scheduling algorithm using new load-balance policy that CPU-intensive tasks have fast cores bias, whereas memory-intensive have slow core bias, how to make Linux kernel asymmetric-aware and reevaluate the load of each core of different type. We propose a light-weight operating system scheduler and successfully solve above problems with negligible overhead and no assumption. We modified Linux kernel 2.6.18 with cpufreq and perfctr driver and finally used Power Now technology to emulate performance asymmetry. Experimental results show that our scheduler outperforms default scheduler and still maintains good fairness, high efficiency and repeatable.
  • Keywords
    Linux; multiprocessing systems; resource allocation; scheduling; Linux kernel; Power Now technology; light-weight operating system scheduler; light-weight scheduler; load-balance policy; multicore architecture; online profiling; scheduling algorithm; single-ISA asymmetric multicore processor; symmetric multicore processors; Benchmark testing; Hardware; Instruction sets; Kernel; Linux; Asymmetric Multi-core Processor; Load Balance; Online Profiling; Operating System; Scheduling;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD), 2011 12th ACIS International Conference on
  • Conference_Location
    Sydney, NSW
  • Print_ISBN
    978-1-4577-0896-1
  • Type

    conf

  • DOI
    10.1109/SNPD.2011.14
  • Filename
    6063563