IETF77 Anaheim

Monday, 22.3.2010, 15:20-17:20

Notes by Christian Schmidt Edited by Yunfei Zhang

PPSP (P2p for streaming protocol) BoF meeting

Chairs: Gonzalo Camarillo (Ericsson), Yunfei Zhang (China Mobile)

Transport Area Director: Lars Eggers (Nokia)

Agenda Bashing

Charter Discussion

Open Issue: Shall we use the BitTorrent protocol as a basis?

Ning Zong: Actual Charter is good. We can use parts of BitTorrent system. Just use it as reference.

Martin Shimaning: BitTorrent as base for tracker protocol. Do we run into copyright issues?

Gonzalo Camarillo: Good question.

David Bryan: BitTorrent is Public Domain Software. Not total optimal for chunk usage. Bittracker.org / Pep3 would be the best solution yet. Flow control and signaling interaction with RTP. Not designed for streaming purposes.

Ning Zong: We can use BitTorrent as basis, but we need arguments, why we do not use PPLIVE/PPSTREAM for this purpose.

Cullen Jennings: Naming of chunks in BitTorrent is an issue, e.g. with usage of time. Do we need something different? There are a lot of questions around.

David Bryan: We made the decision to keep streaming out of the charter. If we use RTP, BitTorrent may not be the best choice.

Gonzalo Camarillo: BitTorrent has some qualities, but not perfect suited.

Open Issue: Media distribution between peers?

Gonzalo Camarillo: Usage of RTP and/or RTSP for this purpose? Should we mention this on the charter?

Matthew Kaufman: I am confused, isn’t streaming media delivery out of scope?

Gonzalo Camarillo: Scope was to start with tracker/peer protocol.

Matthew Kaufman: RTP has not everything you need for streaming.

Henning Schulzrinne: We have RTSP/LEDBAT. We have to identify possible protocols and how to synchronize between endpoints. Both delivery mechanisms should be in scope.

Spencer Dawkins: RTP is not a small thing. A RTP equivalent is hart to do. Is RTP so bad, not to start with?

Gonzalo Camarillo: Should we only promote RTP for this?

Cullen Jennings: See the arguments from Henning. It is a real time problem.

Roni Even: Too early to decide on this. We do not know yet.

Henning: Should be negotiable providing usual interop problem.

David Bryan: Confusing terminology: Real-time vs. time-shifting version.

Spencer Dawkins: Charter should talk about real-time and time Shifting as well.

Open Issue: Distributed or centralized trackers?

Henning Schulzrinne: Two different distributed models are possible, Model 1 - Distributed as replicator, Model 2 - content is offered by several entities.

Gonzalo Camarillo: Working as one logical entity.

Henning Schulzrinne: Model 2 would work according to the P2PSIP mode. Problems are federations and tracker certification. This is far beyond the scope.

David Bryan: Start discussion about distributed tracker. It is a logical function. Do it as P2PSIP. We build this as logical function / start with client/server bases. But keep an eye not to disallow distributed trackers.

Gonzalo Camarillo: We have to nail this down in the charter.

Henning Schulzrinne: Here we have a different solution. The overlay is distributed between peers. The solution is based on a DHT based tracker. A client must not know, about a DHT protocol.

David Bryan: I am thinking about a group of super nodes. Not running a DHT algorithm for this purpose. You can do this with a centralized server. We should not exclude distributed mode out of the scope.

Gonzalo Camarillo: Is search in / out?

Henning Schulzrinne: It is not just a name lookup function. Search protocol of LDAP provides complexity. That’s nothing the group want to tackle.

Cullen Jennings: This discussion has to be reflected to the charter.

Henning Schulzrinne: Globally unique identifiers are needed. Equivalent to an HTTP URL.

Gonzalo Camarillo: Narrowing this is important.

David Bryan: I do not want to use DHT for the search.

Gonzalo Camarillo: Based on a URI, you will find the tracker.

Henning Schulzrinne: Concrete proposal to use URL and a path, a tracker can deal with. The identifier could be something like a URL, not a XML body.

David Bryan: Looking for a swarm ID, must be globally unique.

Henning Schulzrinne: Looking for segment 5 of a certain object. Can put it on webpage or sent it via email.

David Bryan: Should not be an IP address.

Open Issue: Privacy protection

Gonzalo Camarillo: Perfect privacy protection is a good feature to have but not a mandatory requirement.

No objections.

Open Issue: RTP and RTSP (How to initiate connection between the peers, e.g. for file transfer)

Gonzalo Camarillo: This is outside the scope.

Cullen Jennings: What is the p2p protocol for this?

David Bryan: Fuzzy problem, scalability problem, small system with few elements. Query tracker to find a peer and talk with peer about the needed chunks.

Spencer Dawkins: Must this be decided in charter time?

Gonzalo Camarillo: We want to decide it before.

Cullen Jennings: Beware of NATs. In RTP case – a lot of additional things to do. How can we leave this out of scope?

David Bryan: Should it be done here? E.g. SIP extensions?

Cullen Jennings: We probably have to do this here first.

Lars Egers: If the tracker has chunk information, why not use it, e.g. for small systems. Layering is not a good way to implement things.

Gonzalo Camarillo: Interactions have to be included in scope.

Henning Schulzrinne: Which pieces do we have to learn? Perhaps using RTSP extensions for downloading a certain chunk. Do we define a new transport protocol here? Do we define a simple data structure, usable with RTSP or what else? Should we decide here or should we keep this open.

Cullen Jennings: We have no good way in RTSP to select a certain chunk.

Further proceedings:

Gonzalo Camarillo: We will update charter accordingly with the following changes:

- BitTorrent as base: Not the completely suite. Base=yes,

- Media distribution: Streaming and downloading to be supported.

- ICE is in scope.

- Distributed/Centralized Tracker: Search level – out of scope, Identifier, small/big model.

Lars Egers: Assume Centralized design; try hard not to preclude a distributed system at a later stage.

Cullen Jennings: HTTP, SIP, RTSP everything in scope.

Gonzalo Camarillo: The WG will have to decide, based on the requirements.

Cullen Jennings: The charter should state, that we are not willing to do something new or we do not want to invent something new. That’s my concerns.

David Bryan: 99% sure to take something existing. “Will try to reuse...”.

Cullen Jennings: That mean we are not ready to charter a WG for this purpose. E.g. can we use HTTP to transport chunks or do we need SIP? We have made some progress, but this WG could need a lot of time to generate a protocol.

Lars Eggers: That is valid for any WG – have we to decide this before? I think, they can be answered during the WG work.

Cullen Jennings: I want to answer it now, but we can generate a WG nevertheless.

Spencer Dawkins: Should we hom on the decisions?

Gonzalo Camarillo: This makes no sense at this moment. We just want to know, if a WG can be done on this basis.

Roni Even: Depends on the requirements, we did not talk about yet. It is too early to make a decision on this.

Henning Schulzrinne: Goal is to avoid insufficient agreement on the task. A Whiteboard is missing. The detail has to be done in the WG. Does the AD have the trust on the group to work this out within the WG?

Lars Eggers: Is there anything HTTP can do, what RTP (Live stream) cannot do? If this required, than perhaps doing RTP is enough.

David Bryan: There are a lot of reasons to use HTTP for this purpose.

Cullen Jennings: We should go forward.

Gregory Labovitz’s summary:

- BitTorrent, leverage.

- Media distribution: RTP/HTTP open.

- Distributed/Centralized: Centralized, do not harm for distributed.

- Protocols partly

- Privacy: Stays.

Humming / Next steps:

Lars Egers: My proposal is to rewrite the charter and add a Milestone for the WG to decide on open questions.

Ted Hardy: First Charter a WG in this status and then decide, in which time, these open items have to be solved.

Cullen Jennings: Do we need to answer this before we start the WG. First charter a WG, just for answering this open questions and then re-charter. The community should make these decisions.

Gonzalo Camarillo: Sounds funny. Sounds like we will never do this.

David Bryan: Nobody solve every question before starting the work.

Roni Even: These seams to be small questions relate to the rest of the work.

Ning Zong: We should do this in parallel. We should not delay the work.

Gonzalo Camarillo: We have no answers to these questions today. It is a question for the AD, if they are willing to start a WG for this purpose.

Lars Eggers: I am disappointed, that we are further from answering the related questions. Rewrite the current charter and try to come to consensus in the WG. I want to charter the 3 questions as Milestone one. Can the IESG live with this approach?

Cullen Jennings: Is ok with this, as long as community consensus is there.

David Hankins: WG has not talked enough, do not want to charter?

Gonzalo Camarillo: That’s not the case.

Rewrite charter/text, add some modifications, and add a milestone to come to decision about the outstanding open questions.

There was unanimous consensus to go forward of this approach.

Gonzalo Camarillo: We will work on modified charter. We will discuss shortly in IESG and report next steps.

Lars Egers: Add a proposal, how the remaining points should be solved.