- From: Tai Jin <tai@nexus.hpl.hp.com>
- Date: Thu, 26 Sep 1996 10:19:20 -0700
- To: www-talk@w3.org
I hate to followup on my own message, but...
> In either case, there's the additional latency generated by the
> request to the server. Your method has the advantage that the page
> can be displayed before getting a response from the server. However,
As someone else had mentioned, there's also the overhead of server
processing which is nontrivial, especially when using the CGI
mechanism.
This brings up another issue: why can't I put in some kind of meta
data for an inline object which specifies cache parameters? I realize
that caching attributes have nothing to do with markup, but having to
kludge this type of behavior with CGI is not acceptable. It would be
trivial to add these attributes in HTML. It's already there in the
form of the META tag. Perhaps it can be extended with an attribute to
refer to specific named inline objects.
> I'd like to see a status as to how many inline objects I'm still
> waiting for. And the browser must distinguish between animated
> gif activity and download activity and have an option to disable
> the animated gifs.
I'm glad to hear that someone's done this.
> Perhaps a better but less robust solution to the accounting problem is
^^^^^^^^^^^^^^^^^^^^^^
What I meant was more efficient, but less accurate.
> to include in the request a header (e.g., Cache-Hits:) which specifies
> the number of times an object was referenced. Each time a browser
> requests an object, it would include the number of times it was
> referenced from its cache. This counter would be reset whenever it
> got a fresh copy. A proxy would do the same for objects served from
> its cache by returning the total hit count of all the browsers
> requesting that object.
To take this further, I would also suggest another response header
which specifies a TTL for cache hits. That way I can specify that an
object can be served from a cache no more than x times or the expiry
time, whichever comes first, before it must be revalidated. This puts
a limit on the inaccuracy.
...tai
Received on Thursday, 26 September 1996 13:19:41 UTC