On Tue, 16 May 2000, Kenichi Chinen wrote:
> I remember that distributions of `If-Modified-Since' in request and
> `Expire' in reply are not clear in last meeting. Did you decide
> that? and what is decided ?
At the meeting, we agreed that all time-related HTTP header fields will
carry truthful values. For example, Expires: field will contain the
HTTP-exact time of next object modification. Thus, the proxy
(a) can trust the headers and do not have to use
object modification heuristics
(b) must check for freshness of the object to avoid serving
stale content
The above model is arguably better than the current "100% static
content" approach. The model is still not very realistic, of course. It
is probably the most sophisticated model we can have without diving into
staleness issues. As you know, once we try to introduce stale objects
into the workload, we sink because vendors do not want to report that
their caches may serve some stale content and/or cannot agree on the
amount of the uncertainty in the workload.
Object modification and expiration times will be modeled using the
Object Lifecycle model (already supported by Polygraph). The precise
distributions are unknown at this time, but, given the limitations
mentioned above, are probably not that important.
Hope this clarifies,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:12 MDT