Re: Microsoft CARP

On Mon 21 Jul, 1997, Redfern Ian <RedfernI@logica.com> wrote:
>Has anyone got comments on Microsoft's Cache Array Routing Protocol at
><http://www.microsoft.com/proxy/beta/Proxy2BetaSpec.htm>? Their aim
>seems to be to replace ICP with a hashing function that will ensure that
>only one server from a set of peers will hold a given URL. (Note that
>some caches such as Hensa try to make this happen already with .pac
>files).

Some random thoughts.

(note that their URL doesn't display properly in all browsers)

>Their claim is that with deterministic cache selection there is no need
>for any ICP traffic, and clients will be able to pick a peer cache as
>easily as the peers themselves. The proxy array membership table is to
>be available in plain text from a URL.

Hashing the URL isn't a new idea. NetCache seems to have something
similar, though the hashing isn't in the open.

It's a potentially good idea within an organization - if you can
control all browsers and all servers (and I rather liked this idea in
conjunction with a mapping server making the decision, as per the
CRISP "Reduce, Reuse, Recycle: An Approach to Building Large Internet
Caches" paper), but it's no good if you're building a loose mesh as
per eg the JANET experiments.

I don't like the downstream browser having yet another way in which it
can learn about which caches to use (though of course the aim is to
have this implemented in all browsers). Maybe if it can be also be
codified in terms of the .pac file they fetch (which at least is now
cross platform and cross-browser, to some extent) that would help.

What if one adds caches to the cluster though? Where does that leave
you? I'll presume the distribution of URLs across servers is even,
but access patterns to those URLs certainly won't be. You will still
need to 'join' multiple machines to make one visible "server", and
then you've still got to do something to save on duplication between
those multiple machines.

-- jrg.
Received on Mon Jul 21 1997 - 05:26:46 MDT