Re: Last-Modified vs. Expires

From: Henrik Nordstrom <hno@dont-contact.us>
Date: Wed, 06 Aug 1997 22:44:10 +0200

--MimeMultipartBoundary
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: quoted-printable

Duane Wessels wrote:

> Have you seen the Mars Pathfinder home page?
> http://mpfwww.jpl.nasa.gov/
> There are quite a few mirror sites. Why do we need so many mirror
> sites when we have all this wonderful caching technology?

Agreed.

> Lets assume that there were a single Mars Pathfinder web site and
> all the world accessed it throught web caches. Would it have worked?
> I think it would NOT work because with 'no-cache' and IMS, you still
> have a very large percent of requests going all the way to the
> origin for validation.

Yes, but you don't need to turn to infinite caching to get around this.
It is sufficient to simply ignore most of them, but check the object
once in a while, for example twice a month (based on when the object was
last verified with the original server).

I agree that IMS and no-cache is seen far to much today, and there is
several cases where they can be safely ignored, but not "blindly" and
not without some built-in sanity checking..

> I think it would work if the protocol could allow the caches
> to realize that the images (and maybe some of the HTML) only needed
> to be transferred ONCE. New content would be added under uniuqe URLs.

Agreed.

> I also think this is one reason why USENET works pretty well.
> Articles are only transferred once. You can't do an IMS GET
> or a reload for a usenet article.

But that is defined in the design of USENET. If you can get this into
HTTP (i.e. supported by the content provider) then it can be safely
done, but not if not supported 100% by the content provider.

What worries me is that if we add this functionality to Squid without
restrictions, we will see Squid servers that are mis-configured and
caches things that are updated long time ago, and this in a cache-mesh
with parents and siblings and things can go very bad... (one
misconfigured sibling, and you can get a old object that is dated
several months back, when you got a newver copy yesterday from a
parent/server but it has now been expunged from the cache)

My point is that it is always better to play it safe, especially when
you can acheive almost the same goal with very few expenses.

---
Henrik Nordstr=F6m
--MimeMultipartBoundary--
Received on Tue Jul 29 2003 - 13:15:42 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:22 MST