Design and Implementation of an Energy-Aware File System

  • Subject:Power Management
  • Type:Seminar paper
  • Date:22.01.2004
  • Supervisor:

    Dr. Andreas Weißel, Prof. Dr. Frank Bellosa

  • Graduand:Holger Scherl
  • Links:PDF
  • Abstract:

    The development of small mobile hard disks with large storage capacities has changed the application requirements of modern mobile computers and embedded systems significantly. While these devices are mostly battery-powered, all usable power saving techniques have to be integrated to prolong their up-time.

    Most mobile devices suffer from the restriction of small sized memories. It is common that the available memory is exclusively used by the operating system and the running applications. Therefore, in contrast to normal workstation computers, the caching of disk blocks at the operating system level is problematic and not as effective for these devices due to the large memory requirements of that technique. However, existing file systems organize file data and meta data to match performance requirements by relying on the presence of a considerable amount of system memory for caching purposes. As far as hard disks are concerned, valuable battery power can be saved by avoiding disk seeks and introduced latencies due to rotational delays.

    This work examines the effects of different file system layouts on hard disk energy consumption. It is shown that energy efficiency of file systems is heavily influenced by the used data layout and file organization principle when disk caches are mostly not available. Guidelines for a low power file system design are developed. A new implementation of a file system for the Linux kernel based on these recommendations is provided to show that energy efficiency is significantly improvable by adapting the file system layout in systems which do not have the possibility to use disk block caching techniques.


      author = {Holger Scherl},
      title = {Design and Implementation of an Energy-Aware File System},
      type = {Study Thesis},
      address = {Operating System Group, University of Erlangen, Germany},
      month = jan # "~22",
      year = 2004,
      url = {}