Re: [GROW] discuss about draft-luo-grow-ts-use-cases-00

[Christopher Morrow <[e-mail]>]

On Sun, May 29, 2016 at 6:42 PM, joel jaeggli <[e-mail]> wrote:

> On 5/23/16 4:41 AM, luoyuj.gd wrote:
> > Hi everyone，
> >
> >
> >
> >     I'm Jenny Luo，one of the writers of draft-luo-grow-ts-use-cases-00.
> > The draft is mainly about*usecases for traffic steering in operator
> > network*, and we introduce it on 95th IETF meeting in *Buenos Aires.*
> >
> >     Our team received some comments and we appreciated very much for
> > that.   The comments are listed as follows and we would like to discuss
> > them with the relevant experts .  We would really appreciate it if you
> > could offer more description about the comments, so we could understand
> > them correctly.
> >
> >
> >
> > * Comments:*
> > ­1. The requirements are components that sometimes are impossible to
> > implement.Concerns that transferring requirements to implementations are
> > difficult. Fine granularity is a difficult implementation.
> > ­ 2. With the right set of computation someone can do what is needed in
> > a network.Heavily lifting needed to get it through
> > ­ 3. Control system: Add a delay and a random generator
> > ­ 4. What is the different to do it random. Other parties are working on
> > other different constraintorgs.
> > ­ 5. Delaying computation, Implementing at the edge and monitoring
> inside.
> > ­ 6. Not everybody have sufficient resources. Can´t be emulated without
> > resources.  How much spare capacity or overloading in your network? If
> > you can afford serious spare capacity you can aprox. to a good
> > contribution by delaying.
>
> Not sure how these networks are architected, but in my experience
> network cores are rarely over-subscribed unless degraded. interconnects
> between network's may be but political/economic reasons for them being
> so do not see like a great basis for coordination on traffic
> engineering. so that leaves customer facing edges.
>
>
​Even outside of congested cores folk MAY want to do things like: "Hey, for
X pesos less a bit you can send your bits the long-way-around!" and let
users mark their packets for delivery on the long path.

That seems fine, to me... it seems a bit nutty as a configuration
nightmare, but sure you COULD do that sort of gymnastics.​

​I believe I said about this at the mic in EZE, I'd still say the same
thing today.
The document basically wants to outline a set of potentially programmable
methods to say to customers:  "Mark with X, get Y treatment"  and then
deploy configuration that would honor that set of policies to the network.

I don't think this is, particularly, a 'Global Routing Operations' problem,
it's generally an operations&mangement problem to solve.


>
> > ­ 7. How can we schedule flows automatically with fine granularity?
> >
> >
> >   2016-05-20
> >
> > ------------------------------------------------------------------------
> >
> > *Jenny Luo*
> >
> > Data Communication Research Department
> >
> > Guangzhoou Research Institute of ChinaTelecom
> >
> > Tel：[phone]
> >
> > Mobile：[phone]
> >
> > E-mail：[e-mail] <mailto:[e-mail]>
> >
> >
> >
> >
> > _______________________________________________
> > GROW mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/grow
> >
>
>
>
> _______________________________________________
> GROW mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/grow
>
>