Re: [IAB] Call for Comments: "Peer-to-peer (P2P) Architectures"
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [IAB] Call for Comments: "Peer-to-peer (P2P) Architectures"

Gonzalo,
I now see how my initial reading of the document was informed mainly by expectations set by the title and the abstract. I suspect that I might be alone in this. If the title and the abstract were to better reflect the content and to avoid what I read as overpromise, the document would be much improved.
I see you've already changed the abstract to refer to multiple taxonomies. This is helpful, thank you.
I was actually initially confused by the title, "Peer-to-peer (P2P) Architectures". The sort of thing that came to my mind reading that title was a discussion of several architectures that would include design philosophy considerations, main architectural ideas from the P2P world, toolbox, tradeoffs, applicability of techniques, architectural design choices made by particular P2P apps, the motivation of the choices, consequences of the choices, and such. The document does have architectural guidance on when to use P2P as opposed to a server-based arch, but the plural of "architectures" in the title primed me for a discussion of peer protocol choice and design, organization mechanisms to use in the system, and generally something that'd get a server-based protocol jock sufficiently informed about P2P to design working P2P systems rather than to conclude that a given application looks like a potential match for a P2P approach.
If the document were to have a different title, my mindset diving in would be different. If the title more precisely reflected the content I would take the document for what it is instead of expecting something that wasn't intended to be there.
How about "Peer-to-peer (P2P) architecture: definition, examples, and applicability" for a title? Something along those lines would definitely help me get the right idea of the document. 

I've taken the liberty to massage the abstract:
We define P2P, explain how a P2P architecture differs from a client/ server architecture, and provide a non-exhaustive survey of P2P systems and of their taxonomies. We discuss the applicability of P2P and tradeoffs between client/server and P2P and how the best choice depends on the properties and the requirements of the application.
Also, I'd consider pulling out the definition, which currently starts with "We consider" into a paragraph marked with something like "Definition:" at the beginning -- or just the first paragraph of the corresponding section.
And still, I'd suggest making it more clear that, say, BitTorrent, which has trackers, or Skype, which has a website where you can manage your account, are, indeed, P2P systems, despite the fact that they have elements that only provide and don't consume a service. 

Let me know what you think,
-- Stas

--
Stanislav Shalunov
BitTorrent Inc
shalunov at bittorrent.com

personal: http://shlang.com

On Sep 11, 2009, at 3:08 AM, Gonzalo Camarillo wrote:


Hi Stanislav,
thanks for your comments. With respect to including more examples of P2P systems, we received similar comments in the past. Some people thought (like you) that a document having a high number of examples (maybe even an almost-exhaustive list) would be useful for the community. At that point, we decided not to include all those examples in this document and focus instead on general properties and guidelines. However, you are right that such a document could be helpful. During our past discussions, we identified the P2P RG, which was being rechartered at the time of writing the document, as a potential venue to write such a document (we were in touch with the chairs of the RG while writing this document). We realize that there are many potentially-useful documents that can be written in the area of P2P systems and that this document only covers a tiny part of the area.
Regarding a "correct" definition of a P2P system, one of the points the document stresses is the fact that there are multiple definitions in the literature. The document discusses them and describes what they have in common in order to find a common denominator. We believe this approach is more useful than providing our own definition without referencing existing ones.
With respect to the taxonomy, you are right that the document provides several taxonomies and not a single one. Section 4 states that but both the Abstract and the title of Section 4 talk about a taxonomy in singular. We will clarify this point so that it does not confuse readers.
With those explanations for why the document is the way it is, and with the commitment to clean up the single vs multiple taxonomy, would you like to suggest any specific textual edits to this document with these goals? If so, we would happily consider them to the document's next revision. 

Thanks,

Gonzalo


Stanislav Shalunov wrote:
I fail to understand the purpose of this document or of publishing it. 
The abstract promises three things:
* a survey of P2P systems
* a definition of a P2P system
* a taxonomy of P2P systems
I found no traces of a survey, no actual taxonomy, and only an
inadequate definition in the document.
A survey would enumerate P2P systems and describe them.  The document
only mentions two P2P systems of which, one is universally known and
the other is theoretical at present.  Kazaa, or current #2 behind
BitTorrent eDonkey, or the numerous academically interesting P2P
systems are not mentioned, let alone described or classified.
There's a section called "Taxonomy for P2P Systems", but it doesn't
contain any taxonomy, for P2P systems or otherwise.  Instead, it
contains an attempted literature survey of P2P taxonomies.
An actual taxonomy would list the types of P2P systems and their
differentiating principles.  It would be natural to combine it with a
survey, classifying known P2P systems according to the taxonomy you
choose.
An unmarked paragraph four of Section 2 does provide a definition of a 
P2P system.  A discussion follows it, but it's not clear if it's part
of the definition proposed by the document.  The definition is weak
and generic, yet needs to be substantially stretched so that even
BitTorrent, the most common P2P system in use today, fits it.
The document is so bad that proposing specific textual changes is
besides the point.
Here's a sketch of an example of what a more useful document might look like: 
Definition: A distributed system implemented on a computer network to
provide a service is called P2P if it involves a substantial number of 
nodes that, in addition to consuming the service, also provide it.
These nodes are called peers.
Note: P2P systems are often contrasted with client/server systems. In 
the latter, the functions of providing (servers) and consuming
(clients) the service are separated into different nodes; in the
former, the same nodes (peers) can do both. The distinction sometimes 
goes deeper and is reflected on the wire protocol level, where P2P
protocols are symmetrical in nature, with both sides being able to
issue the same set of messages.  This, however, is accidental and not
a defining part of P2P.
Taxonomy
0. Type of service
P2P systems can provide various services, such as file distribution
and sharing; voice and video calling; audio and video streaming,
recorded or live; data backup, etc.
1. Administrative control
P2P systems can be under single administrative control, provide
multiple realms each under its own administrative control, or be fully 
administratively decentralized.
2. Tracker
P2P systems can include the use of a central server that can provide
functions such as peer location and possibly extra services like
content indexing.  Other systems are trackerless and rely on DHTs and
other techniques.
3. Peer equality
P2P systems can contain peers that are elected to perform specialized
functions, perhaps elevating these peers to near-server status, or all 
peers can have the same functions.
Survey
[Skipping the academic systems and only giving a couple of examples.
An actual survey would need many more systems.]
The global constellation of NNTP servers was arguably the first widely deployed P2P system. NNTP provides message exchange and a simple form 
of file sharing.  NNTP was developed before the term P2P became
common, and so the term is not normally applied to NNTP.  However, on
a technical level, if we exclude NNTP clients from consideration for a 
moment, NNTP servers form a P2P system, fully administratively
decentralized, trackerless, and with equal peers.  This system lacks
self-organization and peer discovery, making it a very early precursor 
to modern P2P architectures.  NNTP as a whole, with the clients, is
obviously not a P2P system.
Napster was one of the original P2P systems. It provided file sharing
under single administrative control and with the use of a central
tracker that also provided search.  Peers in Napster were equal.
Kazaa was also a file sharing service under single administrative
control.  It was trackerless and included a form of supernodes
instead.
Skype is one of the most popular P2P systems today and provides voice
and video calling.  It is under single administrative control,
trackerless, and with supernodes.
BitTorrent is one of the most popular P2P systems today and provides
file sharing.  It has multiple realms of administrative control,
usually uses a tracker (but can be trackerless), and has equal peers.
On Wed, Jul 22, 2009 at 12:26 PM, IAB Chair<iab-chair at ietf.org> wrote:
The IAB intends to publish the following document and invites any comments 
you might have:


                  Peer-to-peer (P2P) Architectures
                     draft-iab-p2p-archs-02.txt

 Abstract

 In this document we provide a survey of P2P (Peer-to-Peer) systems.
The survey includes a definition and a taxonomy of P2P systems. This survey also includes a description of which types of applications can be built with P2P technologies and examples of P2P applications that 
 are currently in use on the Internet.  Finally, we discuss
 architectural tradeoffs and provide guidelines for deciding whether
or not a P2P architecture would be suitable to meet the requirements 
 of a given application.




Please provide your feedback before August 15, 2009.

For the IAB,

--Olaf Kolkman
 IAB Chair
_______________________________________________
IETF-Announce mailing list
IETF-Announce at ietf.org
https://www.ietf.org/mailman/listinfo/ietf-announce

Note Well: Messages sent to this mailing list are the opinions of the senders and do not imply endorsement by the IETF.

Note: Messages sent to this list are the opinions of the senders and do not imply endorsement by the IETF.