Cristel Pelsser's blog

Researcher

DECADE BOF

with 3 comments

Purpose: protocol definition for ISP caching for P2P networks.

Motivation: reduce P2P traffic in the ISP network.

Debate: are solutions like webDAV and NFS not suitable, not sufficient?

Comment: requirements as expressed today are too broad. They leave room to a large number of different solutions.

Written by admin

November 12th, 2009 at 12:33 pm

Posted in IETF

Tagged with

3 Responses to 'DECADE BOF'

Subscribe to comments with RSS or TrackBack to 'DECADE BOF'.

  1. Hi Cristel,

    Thinking out loud …

    it seems to me it’s more critical to ask ourselves what will be the requirements in terms of scalability (IO patterns, storage size) which in turn depends strongly on what caching policy would be defined.

    In case of P2P networks, it also probably means – if you want to make it practically working – that user won’t be required to change the config of his client nor the P2P protocol would be changed. The direct implication would be that P2P traffic will be re-routed automatically at the edge of one ISP network towards its cache (CDN infrastructure), right?
    This kind of action is heavy in terms of processing. If you redirect unconditionally all the traffic towards the surrogates, it means that you’ll have a concentrator which should handle all the requests. And in case, you’ll do some heuristics to select the requests which could possibly be in the cache, you’ll put some load on the edge routers. Maybe you could use some kind of layer, à la GSLB, to determine where the request should be sent.

    Whether WebDAV and NFS would be suitable depends – at least – on the aforementioned questions raised. After all NFS is only a layer above the real FS and hardware which are strongly limited by size and IO patterns. Some scale well, some others won’t. You could use behind it an expensive EMC storage or commodity servers with different characteristics. On the other hand, you could use some other DFS like Hadoop(probably not) or Lustre or pick your preferred.

    More, how protect the ISP of not storing/caching prohibited content? Would it be safe for the ISP to cache it?

    Whatever the answers, CDN is an interesting field and caching P2P networks could be a real competitive advantage in the future.

    Do you have some pointers speaking about that subject?
    Do you know some CDN solutions caching some existing P2P networks?
    Are you working on CDN?

    sta

    sta

    8 Apr 10 at 11:46 pm

  2. The DECADE discussion is focused on scalability in terms of network load: The same pieces of data exchanged in P2P networks being carried multiple times by the same physical infrastructure and thus consuming unnecessary bandwidth and router’s resources. They do not consider storage scalability.

    At this stage, it is not proposed that the network automatically directs P2P traffic to the caches. The DECADE community proposes to define a standard protocol that could be used by the peers to store and retrieve content from an ISP cache. It involves modifications of the peers’ software. It also implies a willingness from the peers to use ISP caches. This is only likely to happen if they see some benefit from that, such as shorter download time, availability of rare content, … If this works out, the caches will become part of the P2P network.

    NFS and WebDAV were mentioned from a protocol point of view not from a storage scalability point of view. This aspect was left out of the discussion, mostly because the IETF is a protocol standardization body.

    There was another DECADE BoF at Anaheim. Unfortunately, I did not attend. You will be able to find links to the discussed documents on this page: http://tools.ietf.org/agenda/77/decade.html

    I cannot point to a good reference on scalable distributed storage systems at the moment.
    You’ll be able to find a high level overview of CDNs in this book:
    M. Hofmann and L. R. Beaumont, Content Networking: Architecture, Protocols, and Practice, The Morgan Kaufmann Series in Networking, 2005
    The authors mainly describe the content that is cacheable and how http requests are routed in a CDN.

    Cristel

    Cristel

    9 Apr 10 at 5:25 pm

  3. Hi Cristel,

    of course, it is about network load since we’re talking about CDN stuffs from the ISPs point of view. :)

    I mention storage scalability, simply, because you were asking whether WebDAV and NFS would be sufficient.

    I’m skeptic that a solution which would impose some changes to the user would ever be used at large scale in practice. Of course, such a solution would be the most efficient since designed with cache in mind. But, hey! That would be awesome if it really works out. Of course, if iTunes or any other commercial-file-selling-company could use this kind of solution if global CDN providers/ISPs implements it in their networks. Like I said, I really believe that caching P2P networks could be a great asset in a near future for ISPs.

    I don’t see the point to mention so implementation details like NFS from a protocol point of view. I mean one implementation should be able to store the content in whatever storage type he would like to use. Did I miss something?

    My questions were not related to DFS nor CDN stuffs but about P2P caching in ISPs networks. ;)

    sta

    sta

    9 Apr 10 at 10:01 pm

Leave a Reply