Re: Control on the right to create annotation

[I sent this response to Mark's message on Sunday. It never made it back to me, so I'm guessing it failed to make it out to the rest of the list. My apologies if the problem turns out to have been on my end and everyone else has already seen it...]




Let me make sure I've got this straight: Using PICS as a model means running something very similar (identical, maybe?)  to annotations-sets: Different servers (bureaus, in the PICS terminology) maintain different sets of annotations, and users specify a set of bureaus to subscribe to. This means the annotations system as a whole is fairly impervious to central control or failure. It also means that the system doesn't provide an obvious way to see all the annotations, across the whole system, added to a single page.

Is this right? 

It seems pretty uncontroversial that this would be better than relying exclusively on annotations run as a subservice. Ruling out the subservice option, though, you seem to imply a choice between a centralized implementation and one where annotations are grouped into sets.  This may be false duality. Usenet is an example of a collection of information that's not stored on a central server subject to failure or control, but which still provides a collection of information which is accessible in its entirety (more or less) to a really large percentage of Internet users. I'm not saying that Usenet's the answer- there are a lot of missing pieces in the architecture if you wanted to use it for annotations (most notably, a facility to do fast searches). I just bring it up to point out that collections of information don't have to choose between being on the one hand centralized and complete, and on the other hand decentralized and scattered.

On another note, while I definitely appreciate the problems of scale and control that would accompany a single centralized server (or, for that matter, a few servers...) it's not obvious to me that it couldn't be done. It may well be that the Berners-Lee distributed hypertext model was necessay for the _development_ of the web, but that once the web comes into existence, non-distributed models can work for other web-related services. There are servers out there now that do contain collections of information comparable in size to the text of the whole web: Altavista, Ultraseek and HotBot, for example. And they all claim (although it remains to be seen whether the claim is true) that they'll be able to keep up with web growth.

A centralized server could be a good bit smaller than web-indexes if it stored only the subject lines and urls of comments (in many ways, a very appealling prospect, especially for a centralized model, as it would help distribute the liability for comments: the annotation server would only be holding the correlation between a page and a controversial comment, not the comment itself). My gut feeling is that the real challenge for a centralized annotation server wouldn't be the size of the database, but the number of requests it might have to process.

A centralized model might not be technically incompatible with the goals of annotation (though I'm still unsure whether we all share a clear picture of those goals). Whether it's politically incompatible is another question...

As always, I remain an agnostic regarding the viability of different implementations to support a universally complete set of web annotations. I'm not saying a Usenet-based model would work. Or that a centralized model would. I am saying that it's not certain that they couldn't work. Given how important it could be to pursue this goal, I would want to explore all the possibilities until they either yield fruitful results or prove to be, beyond any shadow of a doubt, unimplementable. 


        - Misha



At 10:42 AM 8/30/96 PDT, Fisher Mark wrote:
>
>Just some quick thoughts...
>
>Annotation is often most useful on material that scared (or "scary") 
>webmasters don't want annotated.  This makes forcing the annotation server 
>to be a subservice of the Web server to be problematic at best.
>
>Since content selection (like PICS) and annotation are closely related 
>services it might perhaps be better to look at having annotation service 
>bureaus, like the label service bureaus of PICS.  Although this does not 
>guarantee that a user would always be able to retrieve the annotations for a 
>given URL, it does decentralize annotation services, making them able to 
>scale much better.  Just as the World-Wide Web would not have been possible 
>if the whole Web needed to be contained in a single server or set of 
>servers, annotations cannot take off unless they mimic the web pages they 
>annotate by having a distributed, transparent, and 
>not-necessarily-fault-tolerant architecture.
>
>The big fault of the Dexter hypertext model is that it requires that all 
>hypertext links can be reached at any time.  This requires a centralized 
>architecture that is incompatible both technically and politically with 
>world-wide distributed systems like the Web and like annotations should be. 
> Perhaps Tim Berners-Lee's greatest idea in creating the Web was that it was 
>OK that a link might occasionally be down or could actually die off without 
>crashing the Web.  A similar principle should apply to annotations.
>======================================================================
>Mark Leighton Fisher                   Thomson Consumer Electronics
>fisherm@indy.tce.com                   Indianapolis, IN
>
>
>

Received on Thursday, 5 September 1996 13:16:49 UTC