This sounds like a discussion for which the beer part is an order of
magnitude more important than the BOF part.

I’m not sure we’re going to get anywhere arguing whether management is a
good or bad thing. 

Can we get back to the discussion about p2psip-diagnostics?

Brian 

From: [email protected] [mailto:[email protected]] On Behalf Of
Henry Sinnreich
Sent: Wednesday, March 18, 2009 3:37 PM
To: [email protected]
Cc: P2PSIP WG
Subject: Re: [P2PSIP] Comment on p2psip-diagnostics

Well, maybe management of overlays works and also maybe it makes economic
sense (for who?).
This would have to be proven however not in papers, but in running systems
as well in market for acceptance.
Too bad the Skype and Bittorrent folks who have proven both do not care to
participate in this debate.
As an engineer, I may be excused for not believing in p2p management, until
I see the measurements and numbers.

> no one knows what happens if the overlay population exceeds some large
number, say >100M, sinces no overlays have been deployed or simulated at
that size - does it degrade >gracefully or become inoperable

My Skype panel shows right now 16,659,464 people on line and I believe they
have 400 million subscribers. 
Skype works exceedingly well for our daily family a/v sessions, coast to
coast.
Does Skype have network management agreements with all the ISPs their p2p is
crossing? :-)

>but depending on user expectations for service quality and willingness to
pay
This alone is worth a good beer BOF.
Let’s plan for one.

Henry


On 3/18/09 11:01 AM, "John Buford" <[email protected]> wrote:

 
2009/3/16 Henry Sinnreich [email protected]
I fully agree, since some naïve folks out there (including me) think that:
1. p2p is self organizing – not managed – unless fixing node software is
called management. 
2. p2p works across the Internet and crosses many ISP networks,  several
times even for any one ISP. The p2p operator may not even be in friendly
relations with some the ISPs and compete with them.
Henry,
 
Current overlays have limitations.  Here's a partial list of problems, I
expect others on this list can add more:
 
- potentially long delays to respond to and correct from network partitions
- inability to detect and correct load imbalance, such as flash crowd effect
- inability to detect and respond to DDOS attacks
- inability to enforce different classes of service for different peers 
- overlay instability at high churn rates, that might be caused by
increases  in the number of mobile peers
- no one knows what happens if the overlay population exceeds some large
number, say 100M, sinces no overlays have been deployed or simulated at that
size - does it degrade gracefully or become inoperable
While there are some research proposals to address some of these problems
individually, it is safe to say that no single design covers all of them.
 
In general as long as there are overlay algorithm performance areas which
can not be automatically detected and corrected by self-organizing
algorithms, then there is a need for management agent(s) to be able to
monitor and intervene.
 
Sure this boundary is likely to shift over time as algorithms get better.
And monitoring the overlay can help the development of better algorithms.
 
Here's a detailed discussion of the limits of today's P2P systems w.r.t.
self-organization:
B. Biskupski, J. Dowling, and J. Sacha, Properties and mechanisms of
self-organizing MANET and P2P systems, ACM Trans. Auton. Adapt. Syst. 2, 1
(March 2007), 34 pp.

Here's a look at how overlay management might work:
 
J. Buford, Management of peer-to-peer overlays, International J. of Internet
Protocol Technology, Special Issue on Management of IP Networks and
Services, Vol. 3, No.1, 2008, 2–12.
 
Bottom line, ideally P2P doesn't need to be managed, but depending on user
expectations for service quality and willingness to pay, there could  be
deployments with overlay operators who provide this for users by managing
the overlay.
 
John
 


_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to