Re: Doubts on Squid Configuration Parameters

squidinfo wrote:

> I am not clear as to how is "hierarchy_stoplist" and "no_cache" different in
> terms of functionality ?

hierarchy_stoplist stops the request from being sent to any parents or
siblings. sort of identical to always_direct, except that it has a lower
priority than never_direct...

no_cache stops objects from being cached in this Squid cache.

> Also "always_direct", "never_direct" and "prefer_direct" have very similar
> behaviours. Could someone give me a concrete example when to use these
> parameters, and in what sequesnce ?

They have different priorities and sligthly different effects

Highest priority has always_direct. If forces Squid to always go direct
to the destination server no matter what the other configuration
directives looks like.

Next priority is never_direct. It can force requests to never be sent
directly to the destination server.

For requests matching neither of always_direct or never_direct the
hierarchy_stoplist and prefer_direct can be used to finetune what types
of requests are being forwarded to peer caches.

Normally Squid tries to go direct for noncacheable requests or request
matching hierarchy_stoplist. "prefer_direct off" changes the behaviour
to first try to forward the request to a parent cache. (note: the effect
of hierarchy_stoplist still applies to sibling caches)

always_direct are typically used if you have local servers which you
want your cache to always go direct to.

never_direct is typically used if you are behind a firewall and must use
a parent proxy.

hierarchy_stoplist is used to make Squid prefer to go direct for certain
request patterns wich often does not result in a cacheable result and
where going via parent caches at best adds latency to the requests
(which is not a good thing).

prefer_direct is typically used if you want Squid to always prefer to
use a parent cache but fall back on direct in case it fails. This is
typically used when it is cheaper to go to the parent than going direct
in which case some people prefer to optimize cost by sacrifying latency.

--
Henrik Nordstrom
Squid hacker