Re: [hybi] Framing Take VI (a compromise proposal)

[John Tamplin <[e-mail]>]

On Tue, Aug 17, 2010 at 8:24 PM, Thomson, Martin
<[e-mail]>wrote:

>  I think that you’ve misunderstood my suggestion as well.  I’m not
> suggesting that we avoid fragmentation (even if there is consensus, and I’ll
> defer to the chairs on this point).  What I’m suggesting is that
> fragmentation can be accommodated as an extension.
>
>
>
> I am absolutely NOT suggesting that we avoid the topic of extensibility in
> the base - to do so would be foolish.
>
>
>
> Repeating Jamie’s suggestion:
>
>
>
>   <length> <opcode>
>
>
>
> Define opcodes for fragmented messages (start, middle, end), then you have
> five opcodes:
>
>
>
>   0 - this is a complete message, and it contains binary data
>
>   1 - this is a complete message, and it contains UTF-8 text data
>
>   2 - this is the start of a fragmented message
>
>   3 - this is the middle of a fragmented message
>
>   4 - this is the end of a fragmented message
>
>
>
> For 2, 3 and 4, you follow that with another opcode - 0 or 1.
>
>
>
> <length> <fragmentation opcode> <opcode>
>
>
>
> You might practice a little judicious optimization, but fundamentally,
> opcodes 2-4 are simply extensions.  An intermediary that doesn’t understand
> just passes these on unmodified (that’s the extensibility bit you have to
> include).
>

Right, and before long we wind up with a long linear chain of extensions
that all have to be evaluated sequentially, and if an intermediary doesn't
know any one of them it can no longer do anything at all with the message,
which I think is a problem.  An intermediary that is just interested in one
extension's data (say a tag that indicates an ingres point to the corporate
network) has to decode all outer extensions before it can get the data it
wants.  Imagine how well routers would perform if IP fields were encoded
that way.

-- 
John A. Tamplin
Software Engineer (GWT), Google