CRTC Items: Flat Rate; Guaranteed Bandwidth

I am going to reverse my decision to not comment on the ongoing CRTC hearings into net neutrality. The reason is that a couple of points came up that I believe are worth some discussion. I am taking my cues from the excellent summary of the first day's proceedings by Frances Munn, as published in Michael Geist's blog.

The first point is about flat-rate vs. usage-based pricing for broadband. Take for example Munn's summary of an exchange between Commissioner Lamarre and Canadian Consumer Groups:

Lamarre went on to address the five percent "high end" users who use more bandwidth than others. She asked whether it was fair for light users to in effect subsidize their usage. The Consumer Groups reiterated that it was appropriate to charge for usage, but expressed concern that prices would go too high too quickly since ISPs remain an unregulated industry. Further, Lamarre pointed out that it is difficult for users to know how much bandwidth they are actually using, and the Consumer Groups agreed that such information would be helpful for users.

I don't understand why it is that people get so excited when the subjects of flat-rate pricing and broadband come up together. This is hardly a novel concept: flat-rate pricing for telephony service, as one prominent example, has long been a CRTC objective for local service, and is now common for VoIP long distance service. Yet there are wide differences among subscribers' use of the telephone, and therefore also in the network costs for the carrier. Broadband is in much the same situation.

There may be some wider variations in usage (a wider Bell curve over the population of subscribers) although this has not been demonstrated. This point was overlooked by both the questioner and the questioned. I have to wonder whether either understands the point. At least Sandvine in its earlier testimony does appear to understand the issue, although it didn't appear to make an impression on the Commissioners:

Sandvine expressed scepticism that monthly usage charging could impact congestion. An economic model that deviates from the simple flat rate 40 dollar/month model would have to have transparent and visible measures of when charges are and are not incurred. Sandvine argued that it could become a powerful model in the long run, but that it would require a change in consumer understanding and acceptance.

The second point is about guaranteed bandwidth. This is a subject I discussed one year ago this month. I was a bit irritated about how it is being misrepresented by Sandvine at the hearing.

...Sandvine argued that their submission was based on how to best manage internet traffic to create an equal space for everyone. They claimed their main purpose is to manage congested internet traffic. For instance, their technology will delay a live streaming video if there is a call attempting to go through the network.

In his presentation, Sandvine's CTO spoke of the need to prioritize traffic according to some carrier-specified criteria - which, not surprisingly, requires DPI - which is about queuing packets and not about guaranteeing bandwidth (what they call "equal space for everyone"). It comes up again during the discussion between him and Von Finckenstein:

First, a service provider has to identify the VoIP packets. All VoIP packets have similar qualities - they are low packet and use relatively little bandwidth. When given priority, the idea is for the VoIP packets to precede non-voice packets.

VoIP packets do not require priority to attain a set level of service quality; VoIP requires a constant guaranteed bandwidth. The reason is that, even when pushed to the front of the queue, there may still not be sufficient bandwidth at all times to assure service quality for this real-time sensitive traffic. It would appear that Sandvine is more concerned with talking their product capabilities rather than the needs of the underlying services; that is, they are promoting what they can do, but not what needs to be done.

Guaranteeing bandwidth is a tougher problem in a packet network that is shared by a multitude of users and applications than it is in legacy networks. The reason is that those legacy networks - circuit-switch telephony, ATM, and long-haul trunks - were designed from the ground up to meet the needs of services that require guaranteed bandwidth. This is also the reason why DOCSIS cable technology includes a telephony channel that is completely separated from that for broadband data.

Guaranteed bandwidth has a cost that is anathema to the traditional data communications paradigm: bandwidth must be reserved whether or not it is actively in use. For example, if you are in a conversation on your wireline telephone, when you are not talking the bandwidth is not available to anyone else. In the same situation with VoIP, packets are not sent and any bandwidth that is therefore not consumed is available to other users and services (or perhaps just other services when it's your home network or your dedicated uplink to the network).

Another way to put it is this: bandwidth on any communications link can only be guaranteed (reserved) for a particular user and/or session by reducing the available bandwidth available on that link to all other users and services. It's a very elementary tenet of network engineering, but an important one. IP networks for the most part do not allow reserved bandwidth and can only manage traffic by adjusting stream and packet priorities by a set of policies. However, none of this guarantees bandwidth.

The schemes that can achieve this on IP networks exist (see MPLS) but are largely not available to the vast bulk of the internet market. Internet services continue to be best effort.