[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [P2Prg] CORE subgroup problem statement

Hi Luca,


Luca Caviglione wrote:
You find your content on a web site. BitTorrent can be possibly classified, from a content localization point of view,
as centralized.
In particular, it is an error to classify BitTorrent as an hybrid protocol. BitTorrent does not have any content localization feature.

<LC>
I would not start a flame about "terminology" :-). However, hybrid protocol is something I don't like in this context. In fact, I used "architectural", rather than protocol. So, I think there are not errors in defying BitTorrent as an hybrid architecture. Actually, there are two architectural flavors overlapped: centralized, from "a content localization point of view" and p2p for the swarm.
Hybrid Protocol suggests functionalities and does not lead to a better understanding of "how the architecture is arranged".
</LC>
<AL>
I don't agree. When you classify architectures as structured and unstructured, hybrid should mean that you have a combination of structured and unstructured.
Otherwise, hybrid is for everything you don't know how to classify. "Other" would have been a better name, but still a weak classification.

It is not correct to classify some protocol based on their localization architecture and some other protocols based on their content replication architecture,
without making a clear distinction in the taxonomy.
If you consider BitTorrent to be an hybrid architecture because there is a central content localization and p2p file replication, then
any other P2P protocol should be hybrid. Instead, using your taxonomy BT should be considered as centralized.

I do not believe I am raising a flame about terminology.
I simply want to point out that the way P2P is considered for now is content localization oriented.
This is only one side of P2P, and content localization and content replication should be uncorrelated. BitTorrent, which is undoubtedly the most efficient general purpose P2P protocol, cannot be classified
using the current taxonomy.
Thus, this taxonomy do not cover all the P2P architectures.
</AL> 


From my point of view, a better taxonomy for P2P is:
*Content Localization
 -Structured
 -Unstructured
 -Centralized
 -Hybrid (?)
*Content delivery
 -Parallel download
 -File splitting  (swarming)
 -Piece selection
 -Peer selection
<LC>
In my opinion, what you suggest is somewhat misleading. I mean, parallel download, file-splitting, piece selection... are techniques for delivering a file over a network and are not strictly related with p2p. The only requirement is that content is replicated among several entities.
Then with your taxonomy, I can describe as p2p "GetRight", the FTP client for Windows. With GetRight you can set mirrors and concurrently download from different FTP servers. Ok, this is an exasperation of the concept, but I think that architectures must be separated from peculiar tricks.
</LC>
<AL>
I am not sure it is possible to abstract the architecture from the technique..Unstructured is not specific to P2P as well. Multicast has been using this kind of architectures and techniques (flooding, expanding ring search, etc.) for a long time.
Structured is not specific to P2P as well, unless you mean Structured=DHT, but in this case you are no more making a distinction between architecture
and technique.
Using the current taxonomy you can already classify protocols that are not P2P.

You are right that my taxonomy is technique oriented rather architecture oriented.
For all the P2P protocols I know, structured means DHT, and unstructured means flooding, expanding ring search, random walk, etc.
I would really prefer a taxonomy using DHT, flooding, expanding ring search, etc. than structured and unstructured.
</AL>
This taxonomy must be refined, but at least, it takes into account both content localization and content delivery.
With this taxonomy, BitTorrent can now be correctly classified:
Content Localization: centralized
Content Delivery: Parallel download, File splitting, rarest first piece selection, choke algorithm peer selection.

<LC>
I think you did not classified BitTorrent. You are only listing some techniques used for achieving a better efficiency on the networked environment.
</LC>
When you classify Chord as structured, don't you mean it is a DHT? Do you have an example of structured P2P architecture that is not a DHT?

From my point of view, DHT is at the same semantic level than piece selection, and Chord is at the same semantic level than rarest first.

I don't know which definition of structured and unstructured architecture you consider, but the one presented in [1] is technique dependent and vague.
Structured is for DHT, unstructured is for other localization techniques.

Regards,
Arnaud.


[1] Lua et al. "A Survey and Comparison of Peer-to-Peer Overlay Network Schemes" IEEE Communication s survey and tutorial, March 2004


_______________________________________________
P2prg mailing list
P2prg at irtf.org
https://www1.ietf.org/mailman/listinfo/p2prg