Re: Proxy problems

On Tuesday 23 July 96, at 20 h 16, the keyboard of stephen white
<steve@adam.com.au> wrote:

> I received the following message from one of my users:
...
> I noted that the proxies don't give you the latest copies of the web
> pages when using Netscape (and I assume the other navigators are the

It is one thing that Joe User complains about "not having the latest
copy" and it is another to discuss it on that list. Semantic consistency
of caches with "original" data is a very old and deeply studied problem
in distributed systems. There is no perfect solution. You cannot have
your cake and eat it. DNS works that way for a very long time and no one
complained.

> proxies. When I checked the ABC transcript list web page, it gave me an
> earlier copy than I had seen the previous day.

It means ABC server does not give good Expires field. Squid and Harvest,
with the time to live computed from Last-modified gives usually pretty
good results: the Web page of a daily paper, for instance, will never
stay long in the cache. May be your cache was not configured to do that
computation (use ttl_pattern) and used default TTL?

> week later when the transcript list changed again. The proxie gave me
> the old copy and didn't give me a clue to ask for a reload. It's

Of course. To do it, the cache would have to make a connection, hence
defeating the whole purpose of caching. See later.

> The way it currently works, there isn't a load time saving or reduced demand
> via Internet because I end up trying reload most times just to see if I have
> the latest copy of the web page.

Yes, if all users were so stupid, we would have to modify Squid to ignore
Reload :-( or to bill them by the byte (with a penalty for TCP SYNC
packets).
 
> The proxie should work by connecting to the desired page via Internet,
> checking the page's date with the date on it's stored copy and if it has
> the latest copy stored, send that to the user, stopping the transfer from

[Good reply from jrg@demon.net (James R Grinter) about latency, something
that Joe User always forget.]

BTW, the real solution to the latency problem is may be to use T/TCP. It
is a transaction-oriented protocol for applications like the Web where
fast response-times for small documents is important. Any news of HTTP
over T/TCP?

http://www.noao.edu/~rstevens/ttcp.html
Received on Tue Jul 23 1996 - 05:33:17 MDT