PANRG

Scion - Update and QA

BT: interim meeting - feedback was to break up scion the work into
component docs, and panrg was happy to accept the work

3 components identified - some close to standard, some need work
next steps are to take the content, put into drafts and publish
independent rfcs - to be worked on in winter, panrg remains forum for
discussion on these topics
panrg not planning another interim, happy to have discussions on mailing
list/next ietf

The multiple roles that IPv6 can play

JL: ipv6 already uses multiple addresses; look into v6 ops, multiple
addresses = hidden limitation from vendors, because multiple addresses
require more resources so need upgraded hardware

JL question; practical testing on fallback speed?

Maxime: quick answer no, but we are looking at doing the work in the
future, look at drawbacks from using multiple addresses, not many people
have experienced practical deployments

Luigi Iannone: are you just looking at developing use-cases in how you
deploy and use ipv6 addresses, or do you want to give advice?
Maxime: no, the purpose is to experiment rather than propose

Ryo Yanagida: have you looked at anything underneath the transport
layer? semantic overloading of IPv6 addresses can affect the network
layer; Maxime: chat more on list

Jen Linkova: unless you go to /64 per host you run into host
limitations; i know transport people don't come to v6ops but maybe
consider it

Maxime: we have a textbook knowledge of v6

Supporting Bottleneck Structure Graphs: Use Cases and Requirements

draft-giraltyellamraju-alto-bsg-multidomain

Speaker: the bottleneck structure model allows to qualify and quantify
interlinks and the relationships between them, this is where it feeds
into panrg; the model also describes how to calculte the effect of
'perturbations' can be things like link capacity fluctuations etc
use cases: wide applications, from routing to congestion control to
network desgin, path selection etc

Speaker: love to get feedback or connect with other wgs

Antoine Fressancourt: question on graph - do you know of research
similar to existing tool, look into it
A: take it offline

Roland Bless ICCRG can be interested in this, research is quite useful
and can solve some cc issues
A: yes, every node has partial information and they are all trying to
converge

BT, as individual: how do I tell whether an arrow in the structure is
uni- or bi-directional (i.e., makes the edge relevant to the
nonpartition of the bottleneck structure)
A: when the link is the local bottleneck for that flow
BT: Talk closer to an iccrg talk, computing the optimality would be
great for iccrg, but for panrg how could we tie the math model to the
signalling we normally discuss

Bottleneck Structure Graphs in Multidomain Networks: Introduction and Requirements for ALTO

draft-giraltyellamraju-alto-bsg-multidomain

Roland Bless:
1: not sure ASes and providers will be candid and want to expose they
have a bottleneck

2: routing based on dynamic metrics leads to oscillations; have you
thought about what you want to do with the outcome - wht happens once
you identify the bottleneck, can end systems choose a different path

A1: the metric does not reveal a bottleneck at an AS; the AS ends up
knowing if it is the bottleck for the path, but other ASes don't know
this - you can reveal this, but it's optional.

A2: this can be used for traffic engineering; what is the expected
performance on the path;

BT: as an individual: i think there is a scalability problem here; does
the computation scale with the number of ASes and edges in the
bottleneck graph inside each AS; seems like it scales extremely linearly

A: the converge complexity is dependent on the bottleeck structure not
the number of paths and ases

BT: can you do something to simplify the substructure inside an AS?

A: for each as, other ASes are simplified and look like black boxes

Database-based Open Resource Service (DB-ORS) Framework, Usecases and Requirements

draft-zhou-alto-dbors-framework

Sabinne, Nokia (contributor to the alto working group): alto tries to
provide guidance for the applications (off-path approach), rather than
steer traffic at the network layer - please look at the alto work
ongoing, where there are proposals to include compute information