Re: [p2pi] Fwd: For those who think "User Fairness/Cost Fairness" is unacceptable...

[Nicholas Weaver <nweaver at ICSI.Berkeley.EDU>]

Assuming for the following that, if you eliminated the 5% heavy users,  
the 95% aren't seening significant congestion...

As such.

On Jun 8, 2008, at 8:25 AM, Robb Topolski wrote:
>
> Since you hold that position, then why do you overlook these?
>
> d) dividing the shared pool of bandwidth by fewer customers so that  
> 5% of the users CANNOT use >50% of the bandwidth

This is overprovisioning which benefits the 5% without benefiting the  
95%, which does not have an economic rationalle (see Brisco's writing  
on the topic for example) in flat-rate pricing.

> e) increasing the shared pool of bandwidth so that 5% of the users  
> CANNOT use >50% of the bandwidth

How is E different from D?

And E is harder if congestion is in the last mile physical media.

And E seems impossible based on the Japanese experience.

> f) dividing the shared pool of bandwidth into smaller allocations so  
> that 5% of the users CANNOT use >50% of the bandwidth

This seems to be "abandoninng statistical multiplexing" mixed with  
"overprovisioning".  For the bulk of the users, statistical  
multiplexing works, and I'd be loath to give it up.


> g) pricing admission into the shared pool of bandwidth so that the  
> price of the largest allocations encourages users to behave  
> economically

Usage based pricing, billed by committed capacity rather than usage.   
(Of course, the ideal is charge for congestion, but that may be hard  
to bill and harder to explain)


Note, that in usage based pricing, ISP's attitude about P2P do a  
complete 180: its a huge profit center: because a user's app can keep  
sucking down and up bandwidth, all through the night (where the cost  
is less but you charge the same), and with the localization, not suck  
up access link bandwidth.

IT, however, kills the P2P companies business model, as they are now  
at least 2x, if not 4x, more expensive to users than conventional HTTP- 
based streaming.

> Perhaps the user has the concurrency problem solved.  Let's assume  
> that they've used their VOIP device and and performed background  
> file-transfers at the same time (because some people have, or else  
> these same people wouldn't depend on VOIP).  If properly managed,  
> shouldn't the network in the middle allow it to work?
>

How is isolating the 5% to fight amongst themselves not doing that?

If you are managing your background transfers RIGHT (eg, capping your  
BitTorrent to a reasonably small fraction, and participating in only  
one or maby two swarms at a time) you aren't part of the problem  
anyway, and probably wouldn't even get to play in the 5%ers "best  
effort" playground if the window of fairness is set to some reasonably  
low time scale (eg, 5-10 minutes).  [1]


> I can see the ads for it now, "...the bulk of our users are mostly  
> happy (...but don't ask what the rest of our users say about us.)"
>
> Why is the definition of fair that "the bulk of our users are mostly  
> happy?" These companies provide Internet to all of us.

Why should the model be "The bulk of the users are unhappy (except for  
the few heavy users...)"?

Why should the 95% of the light users, however, have to pay the  
congestion penalty from the 5% of the users who AREN'T PAYING ANY MORE!

This is a classic externality, and externalities in the network MUST  
be managed if you want a viable, stable network fabric.


Remember, Comcast does NOT claim to do shaping based on protocol  
beyond the protocol-interference of Sandvine.  Yet, strangely enough,  
the complaints of "Comcast Breaking Vonage" really did go away once  
they started blocking seeds and leeches (and ONLY seeds and leeches)

Because it really takes only one or two idiots with their BitTorrent  
cranked to full and running on many torrents to ruin your day on a  
network.  We have seen it.


> Nick, in our current environment, either the incumbant cable or  
> telelephone company is the only real broadband choice.  Lucky  
> residents have the ability to choose between them.  In such an  
> environment, it's everyone's Broadband.   The Internet Access  
> provided by these companies cannot attack one fraction of users in  
> service of the remainder.

At the same time, how can it allow a small fraction of users to attack  
the rest?!?

TCP is flow fair [2], it is NOT user fair.  UDP based crap like Joost  
isn't even flow-fair.

This is, frankly speaking, "Internet Air Pollution Control":  If a  
user uses many flows rather than one, or large non-flow fair  
protocols, should the user NOT be shaped to have the same congestion- 
control behavior as users who are using only one or two TCP flows?

I want user fairness specifically BECAUSE it is "everybody's  
broadband":  I don't want my light flows to be stepped on because some  
neighbor's kid has a dozen Torrents open, because said neighbor's  
traffic will vastly outcompete my traffic, yet he's not paying a penny  
more.



And this might actually be the ideal definition of fairness:  "Methods  
to ENFORCE the network such that every user's congestion response is  
equivalent to that of aggregating all their traffic into, at most,  
four TCP streams under a reasonable queuing discipline".

Would you accept this definition?



Rather, the service you describe, and the service you desire, the  
fully committed rate, without shaping and without bias, is perfectly  
available.  It is marketed to businesses and costs real money.

For the rest of us, statistical multiplexing buys a huge amount.  But  
if you have statistical multiplexing, you must have a way to resolve  
congestion, enforced in the network.

And end-hosts, through the use of recent protocols, have proven they  
are not sufficient.  BitTorrent, as commonly misused, is hostile to  
the user's OWN network, let alone everyone else.  So how does this not  
require enforcement by the network?



> All:  I become increasingly frustrated by being the negative guy --  
> I'm an optimist.  I do believe that Comcast made strides toward more  
> transparency.  But simply advertising that they interfere with some  
> users does not excuse it (see RFC 1087).  When so plainly asked to  
> take a position, the IETF as an engineering body answer Comcast and  
> reject non-Standards conforming behaviors.


[1] You are a great demonstration of why the current model, of app- 
based disruption, shows huge issues and is really evil.  But since you  
are actually tuning your BitTorrent correctly, it wouldn't suprise me  
if in a fairness-based doctrine you'd actually benefit.  Especially  
under the "TCP flow equivelent" definition.


[2] ANd its not even that if you consider the effect of different RTT  
times.

_______________________________________________
p2pi mailing list
p2pi at ietf.org
https://www.ietf.org/mailman/listinfo/p2pi