>>> On 26/03/2007 at 10:40, "Guillaume Smet" <guillaume.smet@xxxxxxxxx> wrote: > On 3/26/07, Henrik Nordstrom <henrik@xxxxxxxxxxxxxxxxxxx> wrote: >> One way is to set up a separate set of cache_peer for these robots, >> using the no-cache cache_peer option to avoid having that traffic >> cached. Then use cache_peer_access with suitable acls to route the robot >> requests via these peers and deny them from the other normal set of >> peers. > > AFAICS, it won't solve the problem as the robots won't be able to > access the "global" cache readonly. As I understand it, the original aim here was to avoid cache pollution (when a robot wanders in, it effectively resets the last-accessed time on every object, rendering LRU useless and evicting popular objects to make space for objects only the robot cares about) - in which case, would changing the cache_replacement_policy setting not be a better starting point? LFUDA should be a close approximation to the result the original poster wanted: anything getting hit only by a robot will still not be 'frequently' used, so although it will be cached initially, it will soon be evicted again. James.
