The main advantages of XMLizing the invalidation are:
- ability to invalidate multiple resources with one message
- extensibility, richer semantics
- overloading HTTP semantics (with a new PURGE method, or by re-using
DELETE) is problematic and may introduce errors (esp. with
interception proxies, etc.)
I would imagine that the payloads of ESI-style invalidation and
WCIP-style invalidation share many aspects, and could be combined;
the means of distributing them is what is different.
Cheers
On Wed, May 09, 2001 at 12:20:40PM -0700, Dan Li wrote:
> See their FAQ on WCIP:
>
> >How does the ESI invalidation specification compare to Web Cache
> >Invalidation Protocol (WCIP)?
> >
> >WCIP and similar solutions try to solve a different problem than ESI
> >invalidation. Although both are a means
> >of communicating invalidations (and possibly other information) from a Web
> >site to caches and CDNs, they
> >have different scopes and applications.
> >
> >WCIP allows a large number of clients to subscribe to a channel, on which
> >they'll get invalidations and other
> >kinds of updates. The primary problems that WCIP addresses are a) scaling
> >to a very large number of
> >clients and b) making the service to all of those clients reliable.
> >
> >ESI invalidation, on the other hand, has a one-to-one relationship; there
> >is no channel mechanism. Instead,
> >an invalidation message passes from the origin server directly to the
> >local cache or CDN. This is more
> >appropriate for sending invalidations to a small number of servers, or
> >when the origin server doesn't have
> >responsibility for propagating the message (such as when a CDN is used).
> >
> >Both WCIP and ESI invalidation might share a common payload (that is, the
> >invalidation message), but
> >WCIP attempts to address much more difficult problems, and is still
> >experimental. ESI invalidation is
> >simpler, more robust, and more appropriate for sending invalidation
> >messages to a known party.
>
> Would the Squid PURGE message (or HTTP DELETE) be good enough for the ESI
> type of invalidations already? or there's some value-add by XMLize this
> type of invalidation?
>
> On the other hand, is WCIP too heavy? do people see a middle ground between
> Squid PURGE-tyle invalidation and WCIP-or-heavier-style invalidation? Or,
> how can we keep it simple & effective, with possible future extensions to
> *much* more value-add?
>
> At 02:15 PM 5/1/01 -0700, Ian Cooper wrote:
> >If you haven't seen it already, you might want to take a peek at:
> > http://www.akamai.com/esi_spec/index.html
> >
> >more particularly:
> > http://www.akamai.com/esi_spec/invalidation.html
> >
> >
> >I'm not too sure what the word "open" means at this point (read the
> >copyright notice).
-- Mark Nottingham, Research Scientist Akamai Technologies (San Mateo, CA USA)
This archive was generated by hypermail 2b29 : Thu Nov 18 2004 - 11:22:57 MST