[ NNSquad ] Re: Is network unneutrality necessarily bad?

[Phil Karn <karn@ka9q.net>]

I agree on several things:

The carriers should disclose all of their policies and mechanisms that may make
their service non-neutral.

The carriers shouldn't under provision their networks.

With DSL, my loop is dedicated to me. (This is not true for cable). There is no
reason I shouldn't be able to use my loop at full speed all the time.

The problem comes with the basic nature of packet switching: it's based on
statistical multiplexing. This is not new; the phone companies have been doing
it from day one. It simply isn't (or it hasn't been) feasible to provide enough
capacity to handle everyone at full speed all the time. And it would be
economically inefficient to do so because it would raise prices for those who
don't use their links at full speed all the time.

That said, I think the carriers are going about this in completely the wrong way
by putting rate limits on individual users and by limiting their monthly usage.
This is a bad match to the underlying technology.

The *right* thing to do is to introduce DSCP (QoS) mechanisms that the *user*
can control. You are right that the user can prioritize his own upstream traffic
so that his VoIP, for example, gets higher priority than his Bit Torrent
traffic. I do exactly this with a dedicated router running Linux and the results
have been dramatic.  I can saturate my DSL upstream link with Bit Torrent
traffic, pick up the phone and make a VoIP call with no impact at all. The Bit
Torrent stream slows down for the duration of the call and then resumes at the
end. It works exactly as it should.

The carrier should give me a similar mechanism for the downstream link, because
I do not control his router. But I must have control over it. In practice
downstream QoS isn't as big a problem as the upstream link because the upstream
links are so much slower.

I don't think we would disagree on any of this. The interesting part is what
should happen beyond my local DSL loop because that part of the network is
shared with other users. Assuming that it is not yet economically feasible to
over provision the shared network as we'd all like (and I don't know that it
still isn't), I think a reasonably fair QoS mechanism could be implemented.

First of all, the carrier should implement a "scavenger" class of service with a
priority lower than default. Users are allowed to present as much traffic in
this class as they want without penalty or monthly limit. Furthermore, there
should be no artificial limits on speed; a link should never be allowed to go
idle as long as there is traffic to send, including scavenger traffic.

Second, the carrier should divide up the available capacity C among the N active
users, defined as those who have sent traffic in say, the last 30 seconds, and
guarantee C/N to each user's default priority traffic. Traffic from any one user
in excess of that amount is downgraded to scavenger traffic, possibly affecting
VoIP so the user has an incentive to mark his traffic accurately.

Basically, each user is guaranteed a fair share of capacity and when one user
doesn't use his capacity, everybody else gets to fight over it.

If a user wants a guaranteed rate higher than C/N, he should have the option of
paying for it.  But there should be no caps on individual data rates or monthly
limits on total traffic even for the cheapest grade of service. Rates should be
based only on busy-time guarantees.

I think this approach is a much better fit to how a statistically multiplexed
packet network actually behaves. Each user gets a guarantee that they do not now
have, yet good incentives are in place to encourage users to do the right thing.

--Phil