News‎ > ‎

Evaluation of Caching Strategies Based on Access Statistics on Past Requests

posted Jan 9, 2014, 5:09 AM by Patrick Poullie   [ updated Jul 21, 2014, 7:48 AM by Corinna Schmitt ]
A paper "Evaluation of Caching Strategies Based on Access Statistics on Past Requests" by Gerhard Hasslinger and Kostas Ntougias, has been accepted for publication at the 17th International GI/ITG Conference on Measurement, Modelling and Evaluation of Computing Systems (MMB 2014), Bamberg, Germany, March 2014.

Abstract: Delivery of popular content on the Internet usually does not rely on a single server but is supported by content delivery networks (CDNs) that reactively store requested content in distributed cache servers. CDNs strengthen the availability and downloading throughput and in addition shorten transport paths caches in the near of a requesting user are preferred.

We study how the cache hit rate as the main efficiency criterion of web caches depends on the request statistics and the caching strategy that selects which content should be placed in or evicted from a cache. Although the least recently used (LRU) strategy seems to be widely deployed in web caches, our comparison in simulations andanalytic case studies reveals essentially higher hit rates for alternatives based on the complete request statistics in the past under realistic assumptions of Zipf distributed user requests.