Re: [Q] 2.0p2 *_direct

Jens-S. Voeckler wrote:
>
> Feasgar math,
>
> I got the following bunch of ACL activated:
>
>...
>
> Now, during very busy hours, both A and B appear temporarily dead (becaus=
> e
> too many ICPs were dropped, answered late, or something -> dead parent
> detected). Then my cache would complain loudly that it cannot select any
> source.
>
> Leaving out the never_direct directive, too many URLs are fetched
> directly. What I want is to fetch the URLs via the parents, unless no
> parent is reachable.
>
> Are there plans for some kind of "never_direct_unless_dead" directive?

I was intending to ask the same question; we now have to pay for
transatlantic traffic caused by requests going direct from our site but not
(at present) for documents fetched via the (UK academic network's) national
cache, and after reluctantly disabling various config settings intended to
improve performance by going direct if the parents were slow, I was alarmed
to see - that around 5-10% of all traffic (POST and HEAD requests, mainly)
was still going direct instead of using the free route via the configured
parent caches.

Thus, Squid's attitude that if you're not behind a firewall and the
responses will be uncacheable, it will not even give you the option of
fetching via the parents, is now costing us real money - but disabling
direct fetches is not acceptable since the cache must be able to fall back
to direct access if the parent cache is down or inaccessible.

From the financial viewpoint (if not the users' - given performance issues
with the national cache), we ideally need a solution for this "now" (or
strictly speaking, a few months ago when charging started :-).

                                John Line

-- 
University of Cambridge WWW manager account (usually John Line)
Send general WWW-related enquiries to webmaster@ucs.cam.ac.uk