Web stuff (was Re: A plea for restraint)

From: Samantha Atkins (samantha@objectent.com)
Date: Sat Oct 14 2000 - 00:35:49 MDT


Eugene Leitl wrote:
>
>Alex Future Bokov wrote:
> > Please direct your comments at how ExtroDot specifically can improve. I
>

One thing that bugs be about most web sites including slash types is
that most of these systems don't have an obvious way to send comments on
the site itself like a webmaster email link. To me it is somewhat rude
and cowardly to not include a feedback button on a site, especially one
that expects thousands of hits.

>
> Nope, I was talking about /. clones in general. I believe it's slick,
> loads fast, and is about as mean and lean as webpages can
> get. Unfortunately, it's still a web page.
>

There isn't anything wrong with web pages per se that a bit of re-think
wouldn't fix. We need imho something at the browser/client (dumb
distinction) end that is more like an interpreter and a bag of very
basic graphic capabilities but little more than that and a fairly simple
scripting language for invoking more capability. What comes from the
server is model (data) plus requests on the views to apply to that data
(more data). The browser constructs the view, loads the model into it
and provides some controller functionality for interpreting user
gestures within particular standard widgets. It does what actions it
can against the data loaded to it from the server (or other peer if you
will) and only talks to the server (or other peer) when it needs
services not available in the browser or the local loadable software
module space. This is much cleaner than having a server that concocts
the entire view, cramming in the model data as best it can with some
ungodly ugliness or other and then sends this thing as a dumb, dead file
that is nearly unreadable to a big fat browser that does its best to
render the thing locally even though most of the brains (what little
there were) have been left at the server end. Almost all client
interaction causes another round-trip to the server for yet another dumb
page whose data is all collected by and at the server and whose
presentation is described and painfully built by the server.

Now I know that today there are a lot of people doing things a little
smarter. Like sending XML data up and having it locally rendered. Some
of them are even using Java or something better to do the rendering.
Which gets a bit better. But too many are using yet more layers of
obfuscation (XSLT, CSS) to get around to doing a half-assed job at
something Smalltalk programmers did well (pluggable views) a long time
ago. I used to subscribe to an XSL[T} list. But it was much too
depressing reading endless posts about people attempting to do simple
programming tasks in something not at all made for it. Why do people
buy some half-ass language or hype of the day and try to make it do
everything?

And exactly why is it that my 650Mhz+ machines with 10s of gigs of
storage are not being utilized for better UI experience, local caching
of data and code fragments and so on? Treat me as a "thin client" when
I'm on a cell phone or my palm pilot but not when I have that kind of
local power. It is a stupid waste. We clog the internet with page
after page to be drawn on machines most of which could calculate their
own pages from data kept in synch with its remote sources with far less
traffic on the net and a far richer user experience and MUCH faster.

The Web as such is not the problem. Falling for hype and forgetting to
think or to dream up better solutions using all the resources at your
disposal is. I have no problem with something that does every
functional thing the web does. I have a big problem with its
dysfunctional aspects and with people and companies literally refusing
to think outside of the box called the WEB. "The customers won't buy it
if we do anything but HTTP for our GUI." "The customer won't buy our
server if it isn't Microsoft." "The customer definitely won't buy our
server if it isn't written in pure Java." Balderdash. Give a business
tools it can use to be many times more effective than its competitors at
a fraction of the cost and they will be lined up at your door. Of
course you have to find the people in the business actually capable of
thinking vs. simply repeating what they read.

> >
> > Huh? Extodot sits on *our* drivespace, not yours. Unless you choose
> > to save a particular posting or article locally. It's *more* efficient
> > on disk space.
>

Let's see. I can pick up IDE space at around 40 gigs for $300 or
less. My time runs around $80+/hour currently. So about 4 hours net of
time wasted putting up with someone else's organization of the data and
waiting for them to render it in the ways their server knows how to and
to deliver it a piece at a time over the net to me vs. spending that
four hours of earnings for enough space to store huge libraries of
information locally. I can see going to other machines (but much more
efficiently) for new fresh information. But going to them to get stuff
that hasn't changed? What for? Disks are cheap, time is expensive. A
remote database/repository is nice for sure but I want my own not little
links to everyone's own web-ified offerings of information. I'm sitting
right on top of a fiber installation at work and on 780K bi-directional
DSL at home. I can afford to download all the original data and updates
I need. On the road I could link to my own server and see things in the
form that I want it.

> I'm a digital pack rat. I don't trust content to be there tomorrow
> when I see it today. Because, like, every third link I click is dead
> like, totally. Rather frustrating, this. Other people are solving this
> by running squid in total web trail mode, me, I don't have that much
> hard drive space. So I grab whatever I deem useful. Because web
> doesn't allow me to save the entire document, I have to resort to
> naked .html, txt, .ps.gz and .pdf
>

Excuse my ignorance but what is a "squid"?

> Actually, I'm running the leanest browser version I know, have a
> browser half life time of a week despite heavy use (abovementioned 100
> simultaneously open sessions) and system uptimes well in excess of a
> month (usually, due to creeping memory leaks (thanks again, Netscape
> programmers) lwhich et ran my swap space (256 MBytes) full, and when I
> forget to weekly kill and restart the browser, the system will sooner
> or later typically crash). It would sure help if I had a gig of core,
> but I'm not that rich.

The Netscape crashes in Linux don't seem correlated with either number
of windows or amount of memory on the machine as far as I can tell.

>
> Opera is not quite there yet when it comes to stability (but it is
> making good progress), nor is Galeon (also almost there). However, the
> problem is principal. You can't render complex content in a
> sustainably stable fashion, using state of the art software
> development. On the other hand, I can rely on the average software to
> render simple content (ASCII) reliably.
>

Actually you can come a lot closer if you keep the data (model)
relatively straightforward in terms of knows sets of meta protocols it
employs and have a stable set of widgets and view thingies that you
might add experimental things to now and then with fallbacks to simpler
ones plus some decent streamable scripting support (both to run
interpreters and embed code and support dynamic configuration). The
rest of the complexity in a mulit-user environment is in concurrency
and transaction type issues and in other goodies like failover,
replication and security. Some parts of transactional processing
(especially long transactions and non-standard models of concurrency)
are research topics. But the simpler things should not be what is
causing unpredictability as is the case largely today. The simpler
things also should not be eating up ungodly amounts of bandwidth. And
most of all the simpler things should be consuming the majority of the
programming talent. When you see that happen something is obviously
wrong.

> I've never liked the web. When the CERN dudes came out with HTML, my
> first thoughts were 1) great, something new which totally ignores
> prior art (both TeX and PostScript) and is not a programming language
> 2) it doesn't do typography? uh-oh. In a few years they'll wind up
> reintroducing basic typographic capabilities, in dribs and drabs,
> until the thing breaks all over the place.

Well, HTML is a bastardized subset of SGML (or was). SGML does give the
real presentation and typography stuff. It would have been really good
if actual display PostScript was used to model a lot of the rendering
stuff and dependably customize it. At the time it came out HTML was a
reasonable compromise for getting people sharing a lot of content over
the internet. But we lost sight of what was and wasn't good about it
and made it a matter of religion to do all things for all people through
HTTP.

>
> So we did have 4 major HTML revisions, now there's XML which is
> falling into complexity morass even before it is widely deployed, we
> have about five different, incompatible widely misused scripting
> languages which further increase browser instability and are a system
> security nightmare.
>

XML is getting made to complex imho. The basic idea is fairly simple.
Except it ignored some simply facts in its early design like the fact
that most data exists in the form of a general graph rather than a
hierarchy. It is taking a really long time for them to get over that
and fix it rather than conveniently ignoring it much of the time. Then
the official standard parsers are braindead in that a 15 meg XML data
stream will blow up into 200 megs or so in memory using most
implementations of DOM (well, under Java anyway). And people got so
hooked on human readability they didn't bother to take XML data
description or DTD and compress the actual data into some predictable
binary format that is more efficient to transmit and decipher given the
meta information and perhaps a bit of indexing at the front. Worse of
all are the people who want to take perfectly good (well, maybe a bit
too bloated) RPC and CORBA and COM pipes and turn it all into XML.

 
>
> > with this point in the first place. Tell me more about your filters.
> > They may already exist as features, or they may be relatively easy to
> > implement.
> >

How about implementing general scripting based filters and let
half-clueful users provide their own filters written in the script? I
would vote for Scheme or Python based scripting myself. So you need an
interpreter and a sandbox and some reasonable API for getting at the
information on your site as information/data.

> > I leave everybody with this question to ponder:
> >
> > Is there an inherent difference between push and pull media, or is this
> > difference arbitrary, artifactual, and potentially transparent to the
> > user in the same way as "static location vs. search query" or "local
> > volume vs. distributed file system"?
>

Push and pull aren't terribly relevant to me. What I really want are
systems that make the walls disappear. I would like to not need to
think much about whether the data, objects, functionality I am dealing
with is on my machine, your machine, all of our machines or what
computer language it is in or what database vendor product and on and on
it goes. I would like to see information/objects/fucntionality helped
to be rearranged and duplicated and migrated to best balance the load
dynamically. I think a big mistake "server" or "website" people make is
assuming they more or less own the information and are the determiners
of how others access it. At most they give some particular views onto
information. But this functionality should not make it impossible to
get to the informationa and bits of the functionality to use them in
other and new ways. I'm probably not expressing that well. Even after
15 years I haven't found a way to say it that will reach most people.
And most of my time gets eaten doing something far less clueful to make
my daily bread.

- samantha



This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:31:36 MST