Re: [PCN] Classification of packets by ingress-egress-aggregate

Hi Michael,

Agree, thank you very much! 
Can you please provide the text that we should include in the draft?

Best regards,
Georgios

> -----Original Message-----
> From: Michael Menth [mailto:menth@informatik.uni-tuebingen.de]
> Sent: dinsdag 10 januari 2012 11:17
> To: Karagiannis, G. (Georgios)
> Cc: Ruediger.Geib@telekom.de; pcn@ietf.org
> Subject: Re: [PCN] Classification of packets by ingress-egress-aggregate
> 
> Hi Georgios,
> 
> the use of the existing PCN variant work from a functional point of view as
> suggested by Ruediger. However, the defined PCN-based admission control
> mechanism may be inefficient in ECMP networks as early blocking is possible.
> 
> -> Section IV.7 in [MeLe12] M. Menth and F. Lehrieder: Performance of
> PCN-Based Admission Control under Challenging Conditions, accepted for
> IEEE/ACM Transactions on Networking, 2012 http://atlas2.informatik.uni-
> tuebingen.de/menth/papers/Menth08-Sub-8.pdf
> 
> This hint should be added as a caveat.
> 
> Regards,
> 
>      Michael
> 
> 
> Am 10.01.2012 08:48, schrieb Ruediger.Geib@telekom.de:
> > Hi Georgios,
> >
> > a statement that tunnels avoiding flow based ECMP routing is the way
> > to go, I agree. I suggest to formulate it that way, as there are
> > tunnels which still support flow based ECMP routing (like IP over LDP
> > based MPLS or Flow aware Transport).
> >
> > We shouldn't mention multiple tunnels as a solution enabling ECMP
> > usage with PCN, if this results in an expectation that we have to
> > provide details how that works.
> >
> > Regards,
> >
> > Ruediger
> >
> > -----Original Message-----
> > From: pcn-bounces@ietf.org [mailto:pcn-bounces@ietf.org] On Behalf Of
> > karagian@cs.utwente.nl
> > Sent: Friday, January 06, 2012 6:58 PM
> > To: toby.moncaster@cl.cam.ac.uk; tom.taylor.stds@gmail.com
> > Cc: pcn@ietf.org
> > Subject: Re: [PCN] Classification of packets by
> > ingress-egress-aggregate
> >
> > Dear all
> >
> > I  agree with Toby. Some solutions are listed in the below draft (Section
> 5.1):
> > http://tools.ietf.org/html/draft-briscoe-tsvwg-cl-architecture-04
> >
> > I think that the solution that will work in PCN is tunelling.
> > In particualr, in networks that are using ECMP then PCN traffic need to be
> tunnelled. In this case the  destination address (and so on) seen by the ECMP
> algorithm is that  of the PCN-egress-node, so all flows follow the same path.
> > Effectively ECMP is turned off. As a compromise, to try to retain some of
> the benefits of ECMP, there could be several tunnels, each following a
> different ECMP path, with flows randomly assigned to different tunnels.
> >
> > Best regards,
> > Georgios
> >
> > ________________________________________
> > Van: pcn-bounces@ietf.org [pcn-bounces@ietf.org] namens Toby
> Moncaster
> > [toby.moncaster@cl.cam.ac.uk]
> > Verzonden: donderdag 5 januari 2012 17:30
> > Aan: Tom Taylor
> > CC: pcn
> > Onderwerp: Re: [PCN] Classification of packets by
> > ingress-egress-aggregate
> >
> > I think we should put it the other way round. In networks that are using
> ECMP then PCN traffic has to be tunnelled or use suitable RSVP. That sort of
> matches with what it says in RFC5559 about tunnelling.
> >
> > I'm no expert, but presumably RSVP must be able to cope with ECMP
> somehow? Otherwise the RESV message wouldn't be sure to return down
> the same route that the PATH message took?
> >
> > Toby
> >
> > On 5 Jan 2012, at 16:03, Tom Taylor wrote:
> >
> >> As GEN-Art reviewer for the SM edge behaviour draft, Joel Halpern has
> maintained that in the absence of tunneling or the use of the aggregate RSVP
> method for determining ingress-egress-aggregate classifiers, routing data are
> inadequate for the purpose. He points out the case where an upstream
> domain uses ECMP, so that the same microflow could enter the PCN domain
> through multiple ingress nodes and the egress node wouldn't be able to tell
> where it came from.
> >>
> >> It appears, then, that the edge behaviour drafts will have to
> >> restrict their applicability to the cases where PCN paqackets are
> >> tunneled or where the aggregate RSVP method is used to create
> >> classifiers. (Even in the latter case, I'm not sure how Joel's
> >> upstream ECMP case is countered.)
> >>
> >> Does anyone have a better idea here?
> >>
> >> Tom Taylor
> >> _______________________________________________
> >> PCN mailing list
> >> PCN@ietf.org
> >> https://www.ietf.org/mailman/listinfo/pcn
> > _______________________________________________
> > PCN mailing list
> > PCN@ietf.org
> > https://www.ietf.org/mailman/listinfo/pcn
> > _______________________________________________
> > PCN mailing list
> > PCN@ietf.org
> > https://www.ietf.org/mailman/listinfo/pcn
> > _______________________________________________
> > PCN mailing list
> > PCN@ietf.org
> > https://www.ietf.org/mailman/listinfo/pcn
> 
> --
> Prof. Dr. habil. Michael Menth
> University of Tuebingen
> Faculty of Science
> Department of Computer Science
> Chair of Communication Networks
> Sand 13, 72076 Tuebingen, Germany
> phone: (+49)-7071/29-70505
> fax: (+49)-7071/29-5220
> mailto:menth@uni-tuebingen.de
> http://kn.inf.uni-tuebingen.de