[ NNSquad ] Fairness doctrines Re: [IP] [with a comment by djf]

["Nick Weaver" <nweaver@gmail.com>]

> - ------- Forwarded Message
>
> From: David Farber <dave@farber.net>

> Even the latest proposals to move away from protocol-agnostic
> management techniques to "fair share" management have been
> criticized as discriminatory against particular users; there is
> currently an active discussion about this between certain
> participants in the IETF's P2P Infrastructure Workshop.  And
> according to some advocates who participated in the P2Pi Workshop,
> "discrimination based on user-history is no better than
> protocol-discriminatory behavior.""

This is a viewpoint that I  don't understand.

One can come up with very clear definitions of fairness:  EG,

"When viewed as a weighted average over a 1 minute period, if N users
are competing for a point of congestion, no user will exceed 1/N +
epsilon of the available bandwidth" (with some smoothing function)


Or:

"During periods of congestion, the network will force all users to
behave as if all their traffic was within a single TCP stream for the
purposes of congestion control"


Or (very close to what comcast is proposing)

X% of the bandwidth is reserved for "low volume" users, who's packets
have a QOS priority.  Whenever the bandwidth desired by all users
exceeds X, the heaviest users (as measured over the past Y minutes
have their QOS marking changed to "no priority" and experience normal
congestion-related packet drops.

As a consequence, lighter users will experience no congestion, and
heavy users will retain the normal "best effort" behavior, but only
amongst themselves.  (Note, X should probably be less than 50%, and Y
should probably be based on some fraction of the median full-rate flow
duration.)


And then, once you define the fairness doctrine, have such mechanisms
enforced by the network.



For those who feel this is somehow unacceptable, how is this more
damaging that usage caps or usage-based pricing (which, eg, eliminates
the ability to perform scavenger services altogether and destroy the
business case for commercial P2P bulk content distribution, and really
hurt streaming-media based distribution)?

How is this more damaging than the current model where misbehaving
users (eg, many-simultaneous torrents, non-responsive P2P protocols)
can disrupt all the other users?


And for those who say "the ISP needs to provision more bandwidth
first", if the stated fairness doctrines provide good service (no
sustained periods of congestion) for >90% of the users, the ISP has
demonstrated that it is properly provisioned in the flat-rate pricing
model.


  [ Might be interesting to see how your average customer service
    rep would explain:

       "When viewed as a weighted average over a 1 minute period, if
	N users are competing for a point of congestion, no user will
	exceed 1/N + epsilon of the available bandwidth" 

    to the typical subscriber upset about their Internet performance ...

         -- Lauren Weinstein
            NNSquad Moderator ]