Decade BOF Minutes
Meeting: IETF76, Thursday, November 12, 2009
Location: ANA Crowne Plaza Hotel, Cattleya East room, 0900-1130
Chair: Rich Woundy, Haibin Song
Jabber: Enrico Marocco, Barry Leiba
Minutes: Lucy Yong

AGENDA (as modified during the meeting)
A.	Agenda Bashing, Background and Overview
B.	Problem Statement
C.	Requirements
D.	Survey
E.	“The Elephant in the Room -- Illegal Content"
F.	Open Discussion
G.	Strawman Proposals
H.	Conclusions and Next Steps

Meeting Report

A.	Agenda Bashing, Background and Overview

Slides are on the IETF website.
Rich Woundy reviewed the agenda.
The goal is to work out achievable scope for in-network storage.
Background and overview
Rich Woundy: this was P2PCaching Bof @IETF75, overview, p2p can leverage in-network storage.

B.	Problem Statement

Rich Alimi presenting
Rich Alimi: p2p content distribution paradigm: scalable, robust, and lots of space for innovation. 
P2P contribute significant traffic. P2P causes stress on network infrastructure. The way to address this is to introduce storage in network.
Problem 1: weakness of existing P2P caches: tight coupling w/ P2P application protocol. 
Problem 2: weak on integration with application. 
DECADE overview, reduce complexity and provide open access, integration with application policies. 
Goal: design standard protocol for diverse P2P applications to utilize in-network storage.
Benefit: avoid complex of existing p2p caching; allow integration with application policies, etc.
In scope: A standard, lightweight control protocol between in-network storage and P2P applications, and a standard, lightweight data-plane protocol for P2P applications to read from / write to in-network storage, and integration examples with one or two P2P protocols
Out of scope: the standard about how p2p app has to use network storage.
Questions:
Ed Juskevicius: Is it realistic that ISPs or others would offer network storage totally free?
Rich W: ISP offer storage accounts today, e.g. FTP accounts. 
Leif Johansson: how does Decade differ from Alto?
Rich A: Alto gives information for peers to use. 
Leif: if I have Alto info, then I don't need this.
Rich W: Alto tells which peers are close, but Decade gives a standard way to store/read data.
Rich A: Alto needs a p2p protocol.
Leif: BT already has this.
Barry: Jabber is lost.
DY Kim: Does decade take care of data ownership? 
Rich A: Yes, a user needs authorization for access to data.
Stuart Cheshire: ISPs use it to reduce traffic. We create it for that purpose. But people may use for other things. Result may not decrease traffic. What will be the incentives to make people use this? What will prevent folks from viewing it as free storage?
Rich A: ISP provides Internet storage. We did not talk about the applications might leverage this storage. Provider could put limits on storage offered to a user.
Enrico M: This work has economic arguments to support it. 
Omer: Applications for it are net meeting-caching, broadcast, and security.
Rich A: Protecting usage of content by other users is also a consideration.

C.	Requirements

Rich Alimi presenting
Data access includes read/write/delete operations.
Resource control allows user to decide how to manage resource allocation for operations performed by remote users.
Authorization is the ability for individual to read, write, and store content.
Data management enables the client to decide time to live for content or to delete content explicitly.
Error conditions handling
David Bryan: One issue that I think also needs to be discussed is if this group will actually make a protocol that handles the actual transfer or we "wrap" some existing IETF protocol that actually does the transfer of data.
Haibin: DECADE can use existing transport protocols for data transfer.
Barry Leiba: What about functions like create list? organize content? atomic move among storage locations?
Rich A: We have this in mind.
Ed: For this to be most useful, this should be near to me as a client. Is there anything in the protocol that indicates how close the storage is to me? Is that relevant to scope?
Rich A: If one storage location is available, then a discovery function is sufficient. If multiple storage locations, alto localization helps.
Stuart: Don't these services already exist? Why we need a new file system? 
Lisa Dusseault: I would like to see a gap analysis. The requirement is very clear. Web server with storage, but missing right tickets.
Lisa: There's a very close match with existing system (Webdav?). What's stopping client applications from using this?  
Rich W: Some P2P applications already leverage existing storage functions, e.g. seedboxes.
Lisa: If a client wants to use caching, why does P2P not use a CDN?
Rich W: Some P2P applications use a CDN to bootstrap a swarm, but that is a different use case. A CDN appears to have additional requirements for in-network storage, which may expand the scope of a potential WG significantly.
David Bryan (Jabber): You should consider using existing transport protocols. [Widespread agreement in room.]

D.	Survey

Ning Zong presenting
Decade system components include discovery and a set of content operations.  
Transparent p2p cache: decade does not leverage traffic redirection schemes such as DPI.
Enrico Marocco:  I want to clarify that this is what we should NOT do, right?
Ning: Correct.
Web cache: HPTPL extension to p2p
CDN: 
NFS: allow client to access local storage.
Draft includes more cases. 
Questions:
XX: Decade should use existing protocols. 
Rick W.: So the future working group would include related technologies in a survey.

E.	“The Elephant in the Room – Illegal Content”

Rich Woundy presenting
Rich W: Focus is on illegal content. User may use this to store illegal content. Illegal content is broader than just copyright infringement.
Illegal content detection and disposition is out scope of this working group. Example: operators are forced to remove or quarantine "illegal content" from storage. But it shouldn’t break applications.
Barry: Summary here is to work on protocol not policy. There are other things belong to policy.
Rich W: Policy has technical influence, which may be good or bad, or change over time. 
Dave Crocker: Realities of world are both social and complex. Don't get distracted by them. There are reasonable and useful ways to distinguish them.
[suggestion to improve the way to address this?]
Peter Resnick: Everyone should read RFC2804; creating a protocol should not help particular laws to accomplish or be enforced.
Leif: Don’t corporate IT departments have similar issues?
Rich A: It is up to user which content should go into network storage, due to possible legal concerns.
Rich W: ISP does not face the same legal issues as a typical corporate IT department. At the same time, the ISP would need to optimize all traffic. 
Leif: Have there been any liaisons to the implementer community?
Rich A: We have been working with the pplive platform; they have interest in using this and integrating with their own software.
Dave B. (Jabber): An advantage to the implementers in that content will be available more widely (potentially when peers are offline) and from better connections. There are incentives for implementers.
XX: In addition to pplive, also thunder and flashget, major p2p systems in china.
XX: It would be good to see that analysis in the survey.

F.	Open Discussion

Charter can be found at http://trac.tools.ietf.org/bof/trac/wiki/decade.
About half in the room have read the charter -- there are perhaps 50-70 people in the room.
Lisa: Channel tap here: form WG and perhaps as a shorter path, suggest writing individual survey drafts with gap analysis and recommendations. Enable divergence in proposed solutions and collaboration. 
Rich W: What are the implications for DECADE?
Lisa: Is there critical mass to work on this effort?
Rich W: How would this relate to the requirements presented before?
Lisa: This sounds like the standards version of the waterfall development model, which is usually suboptimal. The better approach is to consider requirements, pose one or more solutions, then rethink the requirements again. I don’t want Decade to keep coming back for charter revisions.
Lisa: If there is critical mass for people who want to do this, we passed the ambiguity stage.
Barry: The WG needs to have clear milestones. (Few bofs have feasible milestones.)

G.	Strawman Proposals

Strawman Proposal 1: 
Rich A: Service provider provides multiple storage servers. Data locker server can offer multiple accounts.  
Lars Eggert: This is like network file system. 
Rich A: There are differences, such as hierarchical and weighted partitioning. 
Lisa: Partition balancing is not in requirements; was it missed? If so, you may need to revise requirements.
Rich A: Proposal includes access protocol and requirements, including authorization using tokens.
Lisa: HTTP usage security is about use of http and differences between remote procedure inquiry and remote process and remote data access.
Strawman Proposal 2: 
Borjie Ohlman: Network information is a matter of moving from node-oriented to information-oriented network. How to identify information? Info may move from different places and stored in many places. Security is important. API to access any type of object. Organize information according to a model. NetInf naming, flat approach. Netinf architecture.
Lisa: The scope is too broad. It is hard to standardize that many choices. 
Rich W: We want to have a focused scope. The WG may not able to solve your problem.
Song: This is valuable input, thanks.
Craig White: Look for simple ways to consider and incorporate strawman proposals.
Rich W: The NetInf strawman proposal was offered just before the BoF began, so we have not looked it up yet.
Lars Eggert: I suggest looking at the latest NFS specs, as we don't want duplicate work.
Rich W: We will look at it.
Leif: Where should a client use this: at OS level or other? It is not clear in the current requirements.
Rich A: P2P applications would use this. The implementation could be hidden in the client.
Lars: There is a time-to-market benefit for the client to implement a subset of the proposed protocol. It may take a while to put a full implementation into the OS.
Masato: Does this service apply only for ISPs, or can this service be provided by others, such as independent storage providers?
Song: Good question. ISP can provide web hosting today.
Rich W: I don't think that this service is limited to ISPs only. But a CDN provider is not an ISP and may have other concerns. At this time this is an unscoped topic.
Lisa: This is better than <some other> bof. The solution is well described. The feedback and requirements work well here. 

H.	Conclusions and Next Steps

Alexey Melnikov: hum process
First question (“Does the community think that the problem statement is clear, well-scoped, solvable, and useful to solve?”) appears to cover multiple topics. So hum was mixed.

1. Is the problem statement clear? 70/30
2. Is the problem useful to solve? 80/20
3. Is the problem well-scoped? not enough hums. 
Charter language review: “The Working Group (WG) will design and specify a standard protocol for P2P applications to utilize in-network storage. The protocol will include operations for a P2P applications system to (1) store and retrieve data, (2) specify access control on who can store into and retrieve from its allocated storage, and (3) indicate resource sharing policies.”
Tim: The earlier diagram between storage locations meant to me that NFS may be just be missing “server to server” stuff. 
Lars: I am afraid that the NFS solution has been dismissed too quickly and some people assume the existing protocol does not support this.
Barry: We need to figure out what we need to do and an existing protocol may solve this. 
Stuart: Comment on attendance: good stuff, no people do it, so we need new protocol.
Lisa: I believe you can have ISP run storage. There are huge benefits to using it.
Joe Hildebrand: if you use Amazon S3, you've got exactly what we're talking about, in a protocol-non-specific manner. One of the claims for DECADE is to let service providers (ISP) implement one.
Joe Hildebrand: So let's add *another* protocol?
Barry: Or pick one that's close, and extend it.
Lars: If the ISPs just use one protocol, is there a risk for ISPs?
Barry: Google is doing this. There is no standard way to do it.
XX: Storage is not the problem. The problem is control. 
Craig: NFS does not have legal issue. Why does this effort have to address it?
Lucy: The intention is use this as offered service. NFS does not have this aspect. 
Lisa: The reason it takes so long for WGs initially focused on requirements, is because the cycle for requirements drags out for a long time.
Lars: I see no problem with the requirements. The problem is the proposal itself. It may miss more detail on rules and how to do it.

Alexy: Hum: re: working group for gap analysis: unanimous hum
Joe H.: quantify decent, please. [20-ish]
Rich W.: Hands raised for reviewing documents and comment on list: 20
Rich W.: Hands raised for comparison protocols and heavy text contribution: 9