Re: how hint caching works

Sorry about the delay in responding. Like Joe Cooper, I was off
enjoying the U.S. Thanksgiving holiday.

And I was kept terribly busy :-) watching UT Austin's (American-)
football team beat Texas A&M, a grudge match going back 108 years.
 

Henrik Nordstrom wrote:
> Any thoughts how "Hint cache" compares to other local request routing
> mechanisms like CARP in what can be expected real world cache setups?
>
> CARP is quite beautiful in that it has a minimal amont of meta data (only the
> actual formation of the "cloud", not the content) and yet allows requests to
> always hit the correct cache.. but assumes all members of the same "cloud" is
> local to each other unless the requestor also implements CARP routing (in
> which case similar proximity to the requestor is assumed).
>
> The drawback is that CARP is somewhat "centralized/static" in it's nature,
> with a single request path per URL in a given "cloud" formation, not allowing
> it to adopt to different usage patterns.

Yes, in theory, CARP is pretty beautiful. I mean, you talk about your
algorithmic oracle. Can't be beat.

In practice, well, nobody talks about CARP clouds. They talk about
CARP arrays. CARP is designed for a world where there are just a few
huge wonking arrays of caches, fairly far out from the users. And you
can't really link those arrays together atall. They cannot help each
other in any sense. CARP only picks by hash, not by distance to
another cache. If you try constructing a CARP array across the world,
you end up with many wasted queries from San Jose to Xinjiang or or
Tajikistan.

Hint caching is designed for the get-go for a vast number of cheap
caches close to users all over the world. It has concepts of distance
and works to avoid queries unnecessarily crossing the entire world.

-- 
Jon Kay        pushcache.com                      jkay@pushcache.com
http://www.pushcache.com/                             (512) 420-9025
Squid consulting				  'push done right.'