Comments on mogul/cachedraft.txt

Aye! tis a grand document to be sure.

It seems to sumarize the concerns expressed.

The taxonomy of cache typesis very useful to raise awareness of the issues.

Having spent a day reading the archives I see lots of interesting solutions
but keep failing to see the questions. 

cache.
	A OPTIONAL device that may provide certain improvement in some
	metrics such as reducing Latency or network traffic requirements.

	Whether the information came from a cache or an origin server should
	not be discerable by the user (receiving entity, since alot of http
	traffic is now between programs and there may NOT be a user driving
	it).

The terms cache, history buffer, location log etc are useful but I think a
simpler pre-division is required. There are two main types of cache. They
are distinguished by how they are managed. The first type corresponds to
the information stores that are sometimes called caches that are part of
client programs. These are managed mostly by a USER who can flush/reload when
they feel it is needed. The second type is an autonomous cache such as
one running on a proxy server and that does NOT have a user directly managing
it. The second type has to make significant decisions about what may or may not
be cached etc and some of the discussion in this group has tried cover this.


One of the messages "Groupware and Social Dynamics" talks about perceived
benifit. Yet nothing I have seen here actually demonstrates that there IS any
real benifit. Certainly there are major problems that people are trying to 
solve such as the limited bandwidth between USA and Europe and are using 
cache technology to try to solve this but there are other ways.

Some of the mechanisms may have to be done socially. If the demand is there
then provide the resource and charge for it...

Serious imformation providers should explicitly replicate their entities around
the WEB ( as major ftp sites have done).

Caching is not a magic bullet.

The current proposals dont state what the problems are. They do state goals
but don't really justify the goals.

Given that a cache is an OPTIONAL element, there is alot of interaction
between different parts of the HTTP 1.1 spec and caching.

Protocol design is as hard as designing cryptographic algorithms.
There are a few basic mechanisms that can be employed. Other mechanisms just
provide obscurity..

Pete.

-- 
The TIS Network Security Products Group has moved!
voice: 301-527-9500 x123 fax: 301-527-0482
2277 Research Boulevard, 5th Floor, Rockville, MD 20850

Received on Tuesday, 23 January 1996 23:47:11 UTC