Re: [hybi] how do we move forward on agreeing on framing?

[Roberto Peon <[e-mail]>]

On Fri, Aug 20, 2010 at 11:34 AM, Francis Brosnan Blazquez
<[e-mail]>wrote:

> > I don't agree with such a broad brushed statement like that at all.
> > Layers are a convenient way to think about things, but truly layered
> > implementations are almost always inferior.
> >
> > See varghese's _network algorithmics_ for an amazing treatment of the
> > topic.
> > (
> http://www.amazon.com/Network-Algorithmics-Interdisciplinary-Designing-Networking/dp/0120884771/ref=sr_1_1?ie=UTF8&s=books&qid=1282314428&sr=8-1)
> >
> > Protocol designers might use layers as a useful way to articulate the
> > protocol, but it is incumbent on us to consider all the various
> > requirements of different implementation scenarios. That often means
> > layering violations in practice. And that's ok.
>
> Obviously we can't take any rule as inviolable, and obviously perfect
> layer separation is not possible nor desirable...however, what we have
> here is a flagrant attempt to design a complex protocol that was
> expected to only provide transport.
>


Yes, but what does it mean to only transport? Do you want it reliably?
Scalably?
There have been flagrant attempts to do both of these things :)
There have also been attempts to ensure that we have a modicum of security.
These all add complexity and are potentially compelling end results.


>
> It is clear in our particular case, websocket, that it will provide far
> more better benefits if the protocol, as long as possible, keeps its
> features at the transport level, letting other specialized application
> protocols on top of it to provide what the user is looking for.
>
> Without considering the websocket final design, inevitably people will
> layer its favorite protocol on top of it. Assuming that, WG should focus
> on providing features that won't be provided by upper level apps, AND,
> as long as possible do to include features that are already available on
> them.
>


And, hopefully, to reduce overall system complexity by putting complexity in
the appropriate places.
-=R


> --
> Francis Brosnan Blazquez <[e-mail]>
> ASPL
>
>