• DocumentCode
    253181
  • Title

    An energy saving algorithm for variable cache ways

  • Author

    Subha, S.

  • Author_Institution
    SITE, Vellore Inst. of Technol., Vellore, India
  • fYear
    2014
  • fDate
    9-11 May 2014
  • Firstpage
    1
  • Lastpage
    3
  • Abstract
    Set associative caches with variable ways extending to one additional set is proposed in literature. In this model, two sets are searched for tag match during address mapping. Assuming the cache operates in two energy modes - high power mode and low power mode, both the sets associated with address mapping are enabled in high energy mode. This paper proposes an algorithm based on principle of locality to reduce the energy consumption. The last access to set among the two possible choices is chosen as the set to access during current access to a set. If the line is found in the set, it is accessed. If the line is absent, the other set associated with the mapped set is accessed for match. If it is a hit, the line is accessed. If it is a miss, the line is placed in vacant way if available. In case the set and associated set is full, the least recently used way in the mapped set is replaced. The proposed algorithm is simulated with SPEC2K benchmarks. An improvement of 79% in energy is observed with 9% improvement in AMAT.
  • Keywords
    cache storage; power aware computing; storage allocation; AMAT; SPEC2K benchmarks; address mapping; average memory access time; energy consumption; energy modes; energy saving algorithm; high power mode; low power mode; mapped set; principle of locality; set associative caches; tag match search; variable cache ways; Abstracts; Benchmark testing; Average memory access time; energy saving; set associative cache; variable cache ways;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Recent Advances and Innovations in Engineering (ICRAIE), 2014
  • Conference_Location
    Jaipur
  • Print_ISBN
    978-1-4799-4041-7
  • Type

    conf

  • DOI
    10.1109/ICRAIE.2014.6909223
  • Filename
    6909223