Re: [PCN] Redundant aggregate reservations: draft-ietf-tsvwg-rsvp-pcn-03

[<[e-mail]>]

Hi Bob, Hi Tom,



The solution that you are proposing, i.e., use of the e2e RSVP (RFC2205), is in my opinion architecturally and conceptually wrong!

The reason is the following!



In PCN we need a signaling protocol that is able to: (1) to carry information from PCN-egress-edge to PCN-ingress-edge and (2) the carried information needs to be related to an aggregated state, i.e., the ingress-egress-aggregate state.



The e2e RSVP (RFC2205) can only support requirement (1), but is not able to support (2), since the e2e RSVP carries information that is associated with a per flow state maintained at the edges.



Aggregated RSVP (RFC3175) and Generic Aggregated RSVP (RFC4860) can support both requirements, since one of the features that are supported by both is to carry objects that are associated to an aggregated state maintained at the edges. In the context of PCN, the aggregated state is the ingress egress aggregate state.



Best regards,

Georgios



________________________________
Van: Bob Briscoe [[e-mail]]
Verzonden: vrijdag 16 november 2012 21:03
To: Tom Taylor
Cc: Karagiannis, G. (EWI); [e-mail]; [e-mail]; [e-mail]; PCN IETF list
Onderwerp: Re: [PCN] Redundant aggregate reservations: draft-ietf-tsvwg-rsvp-pcn-03

Tom,

The edge behaviour RFCs attempted to abstract over any signalling
protcol, but I think they ended up being biased towards the ITU way
of doing things, rather than an abstraction that encompassed both
RACF and RSVP, which was probably too ambitious.

Is you piggy-backing draft effectively what I am describing -
piggy-backing on e2e signalling?

I don't believe we need any more abstract drafts if that's what you
mean - we need to focus on RSVP specifically now. We could perhaps
re-boot from Francois's original draft.



Bob

At 22:09 15/11/2012, Tom Taylor wrote:
>OK, I guess I see your point. I was too focussed on the calculations
>that are done at the Decision Point.
>
>It makes me wonder if I should revive that piggy-backing edge
>behaviour draft I once started.
>
>On 15/11/2012 4:02 PM, Bob Briscoe wrote:
>>Tom,
>>
>>The RSVP message I'm proposing doesn't say "Never mind the SESSION in
>>this message, I'm related to every flow with the same first hop". It
>>says "These are the marking probabilities for the SESSION in this
>>message". Then the PCN-ingress (not the message) infers that all other
>>flows that share the same aggregate will share the same marking
>>probability, because PCN marking on interior nodes is random and unbiased.
>>
>>It's a subtle distinction, but it preserves the semantics of RSVP
>>messages, without the three disadvantages of setting up an RSVP
>>aggregate that I mentioned.
>>
>>You will have seen from the rest of the message that I have not rejected
>>the concept of aggregation, I am merely saying that the PCN-ingress and
>>PCN-egress can hold the concept internally.
>>
>>
>>Bob
>>
>>At 16:42 15/11/2012, Tom Taylor wrote:
>>>I'm not sure the semantics of the PCN information -- particularly as
>>>it relates to flow termination -- are correct without some sort of
>>>concept of aggregation. Or can you really define an RSVP object that
>>>has semantics "Never mind the SESSION in this message, I'm related to
>>>every flow with the same first hop"?
>...

________________________________________________________________
Bob Briscoe,                                BT Innovate & Design