Meeting notes from ALTO WG meeting, IETF79, Beijing, China
Wednesday, November 10, 13:00-15:00

About 100 attendees

This report has been distilled from the detailed meeting notes taken
by Roni Even, Jan Seedorf and Rich Woundy.

Chairs: Enrico Marocco and Jon Peterson.  Vijay Gurbani in the Jabber
        chatroom.


Agenda
------
[slides, audio]

Enrico presents agenda and status of the WG. Together with other
P2P-related WGs, ALTO is being moved from APP to TSV.

  Peter Saint-Andre: happy with process, will still be involved, not a
  big difference.

  Dave Harrington (as AD): new responsible AD not because of overload
  but more to have all P2P WGs in same area.

Jon proposes interoperability event for ALTO at IETF80/81, want to
start moving from ALTO-like to ALTO, need to start planning now.

  Martin Stiemerling: thinks this is an excellent idea.

  Rich Woundy: thinks this is a good idea, we need to plan this a
  bit.

  Peter: talk to some other folks (e.g. CORE WG) who have dome
  something similar for advice.


Requirements
------------
[slides, audio]

Martin S. presents requirements draft, only one change since last
version: slight chance in "redistribution control" in ALTO reply
requirement.  Some text probably should be moved to deployments
considerations document. Open question is when to finalise this
document.

  Richard Alimi: prefers to keep the requirements draft open for a
  while, e.g. we still need operational experience for rating
  criteria.

  Rich: keep it open until alto protocol draft goes to WGLC.
  Richard Yang: in favour of closing the document earlier.

  Barry Leiba: what is the purpose of the requirements draft if we
  keep it open until we release protocol, that is silly.

  Richard A.: we should at least wait until we have some
  interoperability experience.

  Dave (as AD): like to see requirements published to know what we are
  working against, leaving them open to make them fit to what we
  design is the wrong approach.


ALTO Protocol
-------------
[slides, audio]

Richard A. presents the updated draft. Change summary: protocol
updates, redistribution, IANA considerations. PID naming: enable
hierarchical names with '.' character in names, don't define now.
Service ID (to support servers distributing identical info): use
certificate chains. Cost type registry: needs identifier/intended
semantics/security considerations, require expert review.

  Peter: registries are good, IANA work should be possible with
  reasonable effort.

Remaining issues: provide costs between IPv4/IPv6 endpoints or not?

  Martin S.: keep v4 and v6 guidance seperate.

  Ruediger Volk: asking for costs between address families is
  wierd. Spencer Dawkins: what about dual stack clients?

  Dave (as AD): need to consider what is in the scope of the charter,
  AD needs to watch that the WG works on chartered items. Scott
  Bradner: the charter should not be immutable. Dave: will be
  receptive to discussion for charter changes, may need to involve
  IESG for this.


ALTO server discovery protocol
------------------------------
[slides, audio]

Martin S. presents the updated version of the first individual server
discovery document, previously about third-party server
discovery. Requirements for ALTO are somewhat different than in
geopriv, ALTO has less stringent security requirements than geopriv.

  Martin Thomson: discussion on reverse DNS going on, need to follow
  that.

  Richard Barnes: there is a danger of forged guidance, be
  careful. Martin S.: charter says guidance is only for initial set of
  connections

DHCP for server discovery.

  Martin T.: need to pick one DHCP otpion instead of several ones, do
  not find additional options useful.

Three options for retrieving domain name: user input, DHCP, reverse
DNS lookup.

  Martin T.: it is concerning to shorten the domain name. Peter:
  agrees.

  Randall Gallans: should be in the interest of ISP to provide easy
  ways for apps to find ALTO server, DHCP seems therefore best option.

  Jon: depends on the context, different scenarios may demand
  different solutions.

  Wolfgang Beck: some ISPs use PPP and not DHCP.

  Richard A.: do not forget third party discovery, e.g. by tracker,
  discussion using DHCP, several comments why DHCP will not work in
  all cases and agreement that DHCP cannot be the only option


ALTO server discovery
---------------------
[slides, audio]

Haibin presents the individual server discovery document that was not
presented in Maastricht and not updated thereafter.

  Jon (as chair): we should not have two drafts on the same topic, we
  need to resolve this and need to have only one discovery draft. Roni
  Even: agree that we need to have one draft in the end. Martin T.:
  editors of both drafts could merge for a WG item. Jon: any
  objections to merging the documents?

No objections and agreement from Martin and Haibin to merge the
drafts.


Deployment considerations
-------------------------
[slides, audio]

Martin S. presents the updated version of the soon-to-be deployment
considerations WG document.

  No questions nor comments.


Configuration and monitoring considerations
-------------------------------------------
[slides, audio]

Xianghui Sun presents the individual draft about deployment and
monitoring considerations from operator perspective, intended to be
integrated in the WG document.

  Xen Tien: what is network hop count (slide 8)? Xianghui: any type of
  routing hop count.

  Scott: application download rate is a too generic term.

  Rich: not using DPI for capturing statistics should be
  recommended. Scott: document should not talk about DPI.

  Richard Y.: need to have discussion and feedback from WG what to
  include in the deplyoments document.

  Enrico: document will stay alive for some time, we should have
  discussion on the mailing list.


ALTO and content delivery networks
----------------------------------
[slides, audio]

Richard A. presents the individual draft discussesing integration of
ALTO for CDN with a) HTTP redirect and b) DNS integration. The
document also discusses multi-domain use cases. There is a need to
distinguish endpoint classes in PIDs for CDNs.

  No questions nor comments.


Multi-cost ALTO
---------------
[slides, audio]

Sabine Randriamasy presents an individual draft about a proposed
extension for the ALTO protocol to allow ALTO transactions to use
multiple cost types simultaneously.

  Richard A.: some relation to sensors draft from Cullen and also on
  how they did the parameters for the costing registry in the protocol
  work.

Chairs will like to see this proposal discussed on the list.


Tracker-based peer selection
----------------------------
[slides, audio]

Richard Y. presents a -00 individual draft about challenges for
integrating ALTO in a tracker. One challenge is tracker-scalability,
another challenge is on how to mix application and network
information.

  No questions nor comments.


Xunlei's peer selection mechanism in China Telecom trial
--------------------------------------------------------
[slides, audio]

Kai Lee presents the peer selection mechanism adopted by the Xunlei
client used in the China Telecom trial.

  No questions nor comments.