Adaptive cache management

1  LFU and LRU


2  Both date and frequency

Could setup a cost function considering date of each usage. Establishement and parameters tuning seems hard and computation may be too costly if each cell usage has to be recorded.

Cache management is a kind of allocator. Is the biomimetic approach used for load balancing (another allocator) pertinent for cache management ?

Pheromon marks is an estimation of the cost function ! And as we can't compute future and future is not predictable from past, the random noise introduced by the roulette wheele selector should not decrease performance

3  The experiment

3.1  evaluation protocol

  • Count datasource access
  • apply LRU,LFU,adaptive with various cache sizes.

3.2  test datas

  • Random data
  • Real world data : website logs

3.3  results

4  Conclusion and further works