On Friday 18 February 2005 08:34, Marco Crucianelli wrote:
> On Thu, 2005-02-17 at 12:52 -0600, Kevin wrote:
> > What mechanism are you using to set expire times?
>
> Well, I'm still not sure what I shall use! I mean: should I use
> refresh_pattern!? Or what? I mean, refresh_pattern can let me change
> refresh period based on sire url right? What' else could I use?!
>
when I suggested the choice of two caches, one for small objects and one for
large objects the focus was not on refresh patterns
the goal here is you can use at very first priority the OS and especially the
HD System tuned for serving small or large files. This certainly will not be
possible running two squids on one machine
the second point is that you can use max|min_object_size in order to limit the
file size you will serve by each server. My experienced showed best results
breaking at 512K on modern PCs
third step is to use cache_replacement_policy LFUDA/GDSF accordingly and if
using diskd you may play with Q1 and Q2 what will give you the difference
and to make sense you push from the large_obj_cache with proxy-only set
to achieve this correctly you may or should set additional always|never_direct
for known mpg, avi, wmv,mp3,iso and other so that the small_obj_cache pulls
them really from the larg_obj_cache
Hans
-- _______________________________________________________ Infomatik (18)8112.7007 http://info.matik.com.br Mensagens não assinadas com GPG não são minhas. Messages without GPG signature are not from me. _______________________________________________________
This archive was generated by hypermail pre-2.1.9 : Tue Mar 01 2005 - 12:00:02 MST