DPI: Engineering Solution or... Something Else

I find it sometimes surprising that in all the discussions and regulatory proceedings both here and elsewhere in the world there is little attention paid to the basic economics of DPI for broadband throttling and usage caps. The major ISPs typically claim that they need to take these measures since a portion of their users are causing congestion in their networks due to their extreme practices, especially heavy P2P usage. Following their lead, we end up diving into rat-holes and chasing red herrings in trying to ascertain whether there really is congestion and in considering the privacy implications of DPI.

The more basic question on the congestion question is: which is more economical for the ISP - intervention or adding capacity? It is not that this question is being ignored or avoided by those opposing the ISPs' practices, but rather that the discussion is almost always derailed when it comes up. Since all the data is closely guarded by those ISPs (often, large telecom carriers when it comes to broadband), you can either accept their claims at face value or come up with an independent model of their network and operations economics - which those ISPs will always poo-poo, yet never disclose the true data. Let's explore a qualitative view of the latter - network and operations economics - even while the CRTC seems to lean toward the former - trusting the carriers - in their deliberations.

First, there is the matter of capital intensity. As reported here, Andrew Odlyzko makes the claim that there is no need for increased investment to achieve higher capacity. (He goes further and says that internet traffic volume growth has slowed somewhat, but let's leave that aside.) This seems very believable on its face . The long term capital budgets of all the large carriers is fairly steady over time as a proportion of revenue. Even so, they have built massive wireless and broadband networks from scratch, and which continue to increase in capabilities and universality.

Part of this is easily explained: if you want to buy a piece of equipment today you will be hard pressed to get one with 56 kbps line speed - it's obsolete and only available on the used market, if at all. If you go out to buy equipment, like it or not you will get the latest and greatest, and you will pay about the same price as in the past for its less-capable predecessor. While time may not cure all ills, it will result in higher-capacity networks with no change in capital intensity on the part of the ISPs. We then only need to consider whether this natural capacity growth is keeping pace with user demand. I have no hard numbers to present in this post, however all indications I've seen is that user demand has yet to overwhelm network capacity in either the access or core. It is also naturally limited by user connection speed, whether DSL, DOCSIS or wireless - if you have 3 Mbps DSL, you can't use 4 Mbps - and the ISPs charge preciously high prices for higher-speed tiers. They are getting additional revenue from the high-usage subscribers they decry as a blight on their profitability.

For my second point, we must understand that capacity is not free; if there is congestion it will cost money to alleviate it, however it is done. The only alternative would be to modify user behaviour, but that is unlikely without draconian measures such as raising prices or limiting availability - neither the ISP nor the user benefits from that. Therefore there are two technical strategies that can be performed in the network to alleviate congestion: increase capacity or reduce usage. Which is the more economically attractive? Adding capacity is technically easiest: just add more of the same, whether it be more or faster data connections within the access and core networks, and to peer networks, as subscribers are added and as per-user consumption increases. They just need to add up the prices of the new assets, write off the assets being replaced (old cards and boxes) and figure out how much operations costs change for the projected life of the asset. Since capital investment for mature technology is usually in the range of 25% to 50% of total expenditures, this can be modeled with some confidence.

Now let's consider DPI as the technology to reduce usage. Have you ever tried to sell a new box, a piece of the network infrastructure, to a carrier? I have, and I can tell you that it isn't easy. You'd better be a very athletic salesman if you hope to jump all the hurdles they put in front of you on the path to your goal. The prospective customer will roll his eyes, grimace and otherwise look for any and every reason to avoid doing business with you. Depending on the specific technology, the carrier may demand that you partner with an existing supplier and convince them to incorporate your technology within their box, just so the carrier doesn't have to add a new box to their network.

DPI includes a box (network element) of this latter classification. Since it is in series with the users' data it becomes another point of network failure which must be accounted for in network engineering. It must also be integrated with the carrier's OSS (operational support systems), including provisioning, maintenance and accounting. On top of this, staff must be trained for the new equipment and services, and service requirements must be designed, tested in a lab, and deployed, with the possibility of backing out if there are problems. Not only that, since it manipulates user data there is a risk that user services will be (unintentionally!) degraded. As can be seen, there are good reasons for the carrier to be extremely wary of introducing new network elements - there must be a compelling business reason. Unlike the costs associated with existing technology, the capital costs for something entirely new is typically only about 10% to 15% of their total costs - the box itself is essentially free.

This is why I am dubious that the ISPs deploy throttling DPI to solve network congestion. There is an easier solution, and likely at least as economical in the long run: add capacity. As an engineering solution to congestion it would be difficult for the decision-makers at these companies to approve the necessary budgetary requirements, especially when you also consider the high risks of the public relations strategy in which they are currently engaged. I therefore conclude that DPI has not been deployed as an engineering solution to an engineering problem.

If not an engineering solution, the ISPs' drivers must come from marketing. This involves many possibilities, including new service revenue, innovative service opportunities, competitive positioning, and so on. Everyone has an opinion on what, exactly, they are up to, but since they are less than forthcoming about their reasons (as should be expected), there are many suppositions. These may or may not include an evidence trail. Some of the possibilities came out on Friday at the CRTC hearings when it became clear that DPI was not used solely for congestion control but for other purposes, including, but not proved, anti-competitive behaviour.

These hearings are still more politics than substance, even with the revelations that have come to light. The question to my mind is whether the evidence will convince the CRTC to rule based on the evidence instead of political policy. We won't learn the answer to that question for some time yet.