Re: [vwrap] Consensus? What exactly should be in the protocol

[Cristina Videira Lopes <[e-mail]>]

All the limitations that you mention about the Web architecture not 
being enough to support virtual world applications have been muted by 
HTML5. Additionally, CORS now allows for true client-side mashups.
But even without these two things, you can build non-web-browser clients 
that follow the general principles but that do special things for the 
real-time updates -- basically, the general concept of 
JavaScript+WebSockets done in whatever other way you like: different 
programming language, different protocol,...

The really important architectural principle, though, and one that is 
unlikely to be let go, is that the use of WebSockets, the data formats 
that flow through them, and the use of CORS, are decisions that pertain 
to *each* virtual world application, it's not something that is imposed 
on all VWs by web standards -- it comes as JavaScript sent by the 
server! They are *implementation options* -- very valid options, I must 
add, but options nevertheless. What you are trying to do here is to 
dictate that all virtual world applications MUST use some protocol for 
renderer -- server interactions, down to the data formats, and MUST use 
capabilities for mashing things up, or else... they can't interoperate.

You can dictate that. But then this will be completely irrelevant in a 
couple of years when WebGL is actually usable or when Google finishes 
their virtual machine for running safe native code on browsers.



David W Levine wrote:
>
> So, of course we're building in the web space. I hope nobody is 
> denying that. In fact if you look at everything described in VWRAP is 
> starts with an assumption that most services are delivered as REST or 
> REST like services. I think its safe to say that the people who have 
> been discussing this for over two years are aware of Roy's work, and 
> have thought about how REST applies to virtual worlds. REST represents 
> a lot of thinking about how the web delivers content, and in 
> particular why not to turn the web into a
> distributed object model, or a shared state model, but rather to 
> leverage the observed successful patterns of the web in managing 
> distributed programming problems.
>
> But.. (There is always a but) The very core thing that a virtual world 
> does doesn't fit terribly well into the mainstream web model. The 
> heart of a virtual world is delivering (and Morgaine's phrase serves 
> very well here) a visual mashup of things to users 30-60 times a 
> second, updating continually to reflect the input of the physical 
> simulation, any user
> inputs, and any scripted inputs. Our core problem is taking in the 
> inputs, deriving the new state and sharing it out to the users. This 
> isn't really what the web has historically done. The fact that it 
> isn't, that there are some really interesting distributed system 
> challenges at the very heart of this, is part of its technical appeal 
> to me.
>
> Life is made harder by the fact that the virtual space is being 
> constantly asked to accept new things to deal with. Every time an 
> avatar arrives it brings a set of stuff
> which has to be melded into the scenegraph. Again, we all know this. 
> Rezing an avatar means adding a bunch of new content to the virtual 
> space, and it means pushing
> it back out to all the observers.
>
> In the traditional web you go to a URL, you do a get, and you get 
> handed a huge slab of stuff to render.(some of which may require 
> fetches, plugins, etc.) In the more dynamic 2.0 style stuff, the stuff 
> you get may include dynamic elements which fetch and update more 
> stuff. In the virtual worlds space, we bring to to a fever pitch. we 
> take inputs from all the present users, from a simulation, including 
> the scripted changes within the simulation. We then turn around and 
> want to show this to the user.
>
> How do we present this to the user. Well, we currently use Linden's 
> UDP/http/longpoll tangle. Fine. But. how could we do it?
>
> We could create a video stream and stream it. (WHich isn't very web 
> page like at all, but has some nice properties)
> We could do something like OnLive where we would create a very 
> tailored stream and deliver it to a client with very specialized 
> coupled inputs (And life with a lot of
> constraints and again isn't very web page like)
> We could send a stateless update every frame for the client to render 
> (Well, with ulimited bandwidth and processor power)
> Or.. we could do what we currently do, just cleaner. which, roughly 
> speaking is send down initial state and then send down a series of 
> updates to that state. Woah, not
> exactly a traditional web page. Worse still.. where do we post the 
> inputs from the client to the world?
>
> At the same time, we also get to ask "How do we get all the "stuff" 
> into the region. In Linden's world, the answer is easy. They use a 
> proprietary protocol and fetch it from
> their creaking central servers. In OpenSim, a similar answer obtains. 
> And for added pain which we have all shared, the current set of 
> clients push all the stuff related to
> the user via the region.
>
> VWRAP attempts to describe nothing more than a set of REST web 
> services which represent the region and the services. It attempts to 
> leverage what's been learned from REST, and Linden's system, and in 
> fact OpenSim, to describe a simple, extensible set of services which 
> can describe: Regions, Auth services, how to rez and unrez avatars, 
> how to (when we get some writing down) fetch and manipulate assets, 
> inventory lists and so on.
>
> What you end up with is built deeply on web principals, but not a web 
> page, but mostly because a virtual world is not, at its heart a web 
> page,but a set of services collaborating to
> share state in a pretty unusual way.
>
> - David
> ~Zha
>