On Wed, 20 Apr 2005, Michael Scheibel wrote:
Why is LRU still the default replacement policy; others seem to perform much better in terms of hit ratios.
Because LRU performs reasonable, and is very lightweight for the proxy to maintain.
"The results indicate that the heap-based implementation of the GDSF and LFUDA policies consume less user and system CPU time than the original LRU replacement policy in Squid 2. Interestingly the heap-based LRU implementation also consumes less CPU time than the original list-based LRU implementation."
This was probably due to a small hit ratio optimization done in the old lru algorithm in earlier versions of Squid. This is no longer done since the refactoring of the removal policies in Squid (Squid-2.4 I think).
Regards Henrik
