some HLL comments

Jecel Mattos de Assumpcao Jr. jecel@lsi.usp.br
Tue, 7 Mar 1995 18:32:06 -0300


On Mon, 6 Mar 95 13:19 EST Raul Miller wrote:
> Francois-Rene Rideau
>    On netscape, the "reload" icon should flush this cache; but I'd
>    recommend you have a local copy of the whole Tunes distribution,
>    and then use the GNU patch utility to update it.  Patch 0.0.0.9 is
>    just out !
> 
> Nope.  This is one of the issues you have to deal with in a
> distributed system: computational integrity with distributed data.
> [And my "favorite" hobby horse: many people with computer science
> training are used to thinking in terms of "atomic" operations which
> maintain computational integrity, but distributed systems need some
> form of instancing instead.]

This is what I spent most of 1993 and half of 94 think of in the Merlin
project. I still can't handle replication very well, but I am working
on it.

> The problem here is that Jecel does not have direct access to external
> HTTP servers.  Instead, an http query is handled by a local server
> which caches query results for some period of time.  During that time
> period, the locally cached copy is used.  This saves communications
> bandwidth when compared to fetching the same text from remote sites
> every time someone looks at it.

The problem is that the local server has lots of space for the cache
and tries to be smart ( but fails ) at guessing when a page might have
been changed or not. I can bypass the local server if I use Mosaic
rather than Netscape, but it isn't a very good solution at 1200 bps.

> There's several ways around this caching mechanism.  The "right way"
> is to issue a unique id to web pages when they change.  Typically,
> this means incrementing a version counter (or generating some sort of
> hash code).

The problem goes beyond caching. The browser shows links you have read
in a different color from those you haven't seen yet. If a page changes,
the browser still tells you that you have already looked at it. Ted
Nelson's Xanadu didn't have this problem, I think.

> Another workaround is to fetch enough data from the web to flush the
> cache of the server.  However, this isn't necessarily easy (or even a
> good idea).  For instance, think about fetching several gigabytes of
> information (for a server with a big local disk) across a 14.4 slip
> line...

14400! You're lucky...

-- Jecel