[4489] in SIPB bug reports
Re: Mosaic
daemon@ATHENA.MIT.EDU (yandros@MIT.EDU)
Sat Mar 19 16:41:32 1994
From: yandros@MIT.EDU
Date: Sat, 19 Mar 94 21:40:56 GMT
To: sommerfeld@apollo.hp.com
Cc: bert@MIT.EDU, bug-sipb@MIT.EDU
In-Reply-To: <9403171723.AA26638@MIT.EDU> (message from Bill Sommerfeld on Thu, 17 Mar 1994 12:23:00 -0500)
HTTP does not include *any* support for cache coherance -- either
callbacks or TTL -- and this is a major shortcoming of the protocol,
which prevents it from *really* scaling up in the same way that AFS or
DNS scale up; any "TTL" that an HTTP client assumes is going to be
wrong in at least some environments, and an infinite ttl (which is
what Mosaic uses for inlined images) is Just Plain Wrong.
I had this conversation with Ted. He assumed, as did you, that HTTP
documents had no TTLs. You are both completely and utterly wrong.
There are already caching servers running *right now*. Just two days
ago I saw an announcement of a group that wants *anyone who wants* to
use their single caching server to see how much of a beating it can
take. Clients are getting more and more intelligent about caching all
the time. There's a group in the UK writing an almost-user-level
caching server (since most clients have the ability to use a cache
server built in, this is easy).
It is true that most things don't use TTL's yet.
It is not true that they'd be hard to add or even that they don't
exist yet.
At this point I resist the temptation to say `RTFM, bonehead'. But
only barely. :-)
chad
grumble grubmle people complaining who don't know what they're talking
about grumble grumble ashould be doing 6.170 grumble grumble :-)
