Re: [core] wglc: coap caching and observe, caching underspecified

Hi,

On 04/20/2012 02:23 PM, Thomas Fossati wrote:
> Carsten,
> 
> On Apr 20, 2012, at 2:51 PM, Carsten Bormann wrote:
>>> This thread seems to be leading towards a separate bit in the option that indicates if a proxy needs to understand the option or not. Not sure what I think of that idea but it would be one possibility to consider. 
>>
>> I would like to see a real-live example for an option that is neither critical nor elective, first.  We have had this discussion for a long time, but nobody ever came up with an option that would *need* that third/fourth state.
> 
> the issue here is that we need a way to distinguish end-to-end from hop-by-hop semantics in options, otherwise proxies can't cooperate in making CoAP networks scale.
> 
> I don't mind if we want to do that with an explicit bit, or by some implicit rule -- as I proposed previously in this same thread -- but we need to do that.
> 
> The current caching rule set treats an intermediary the same way as an endpoint, forcing a flow to break at a forwarder whenever some end-to-end semantics is not understood.  And this is plain wrong.

So I've no real opinion as to whether this extra bit is
needed/useful or not, but if its added then it'll need
to play well with how DTLS is done.

Be a shame to accidentally muck things up embedding some
hop-by-hop option into something secured end-to-end.

I would guess that any solution here that isn't "don't
use hop-by-hop options if you need (e2e) security" will
be fairly complex.

S

> 
> We don't know how much CoAP will need to scale, but embedding such a limitation in the core protocol is something that we could regret.
> _______________________________________________
> core mailing list
> core@ietf.org
> https://www.ietf.org/mailman/listinfo/core
> 