Re: how hint caching works

On Saturday 24 November 2001 06.13, Jon Kay wrote:

> 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.

True. CARP can only build trees of clusters, not truly a cloud or even mesh.

But from political reasons or whatever, it is quite hard to get cooperation
across administrative boundaries, making people looking into caching to
mostly look at their own network domain, making me thing that CARP fits quite
well in the most common network conditions for caching.

You have to compare the cost of maintaining a distributed cloud in both
traffic and commercial conditions, to the cost of simply extending the local
part to be large enough. Provided ofcourse your user population is large
enough.

For pushing, CARP is mostly worthless. I am only challenging the pull caching
side of your claims for realistict real life network conditions, where I
think the two approaches can be compared.

Regards
Henrik
Received on Sat Nov 24 2001 - 02:06:02 MST