Deep Packet Inspection and Mobile Discrimination

Throughout the 2009 Canadian Telecommunications Summit presenter after presenter, and session after session, spoke to the Canadian situation concerning growth in mobile data. In essence, there is a worry that the wireless infrastructure cannot cope with the high volumes of data that are expected to accompany increasing uses and penetrations of mobile technologies. Such worries persist, even though we’ve recently seen the launch of another high-speed wireless network that was jointly invested in by Bell and Telus, and despite the fact that new wireless competitors are promising to enter the national market as well.

The result of the wireless competition in Canada is this: Canadians actually enjoy pretty fast wireless networks. We can certainly complain about the high costs of such networks, about the conditions under which wireless spectrum was purchased and is used, and so forth, but the fact is that pretty impressive wireless networks exist…for Canadians with cash. As any network operator knows, however, speed is only part of the equation; it’s just as important to have sufficient data provisioning so your user base can genuinely take advantage of the network. It’s partially on the grounds of data provisioning that we’re seeing vendors develop and offer deep packet inspection (DPI) appliances for the mobile environment.

I think that provisioning is the trojan horse, however, and that DPI is really being presented by vendors as a solution to a pair of ‘authentic’ issues: first, the need to improve customer billing, and second, to efficiently participate in the advertising and marketing ecosystem. I would suggest that ‘congestion management’, right now, is more of a spectre-like issue than an authentic concern (and get into defending that claim, in just a moment).

Before diving into this issue, however, I want to be up front with you: much of what I’m going to be referencing in this post is from last year (2009) because I’ve had a pile of stuff I’ve meant to write about but haven’t had the time until recently.[1] I don’t think that this invalidates what I’m writing, but think that you deserve to know ‘when’ quite a few of the links will lead to.

Let’s begin with the issue of mobile ‘data hogs’. Last December, AT&T started making noises that iPhone users were ‘data hogs’ and some kind of extended monetization strategy was required. An economic solution was preferred on the basis that (a) it would generate revenue for AT&T; (b) economics are typically seen as a stellar way of dissuading popular use of expensive new and emerging technologies. In AT&T’s specific case, of course, there were questions about the degree of actual mobile 3G investment but, more significantly, the question of whether ‘data hogs’ were actually the problem.

You see, contemporary smart phones, such as the iPhone, iterations of Android, and Palm Pre, have been designed to maximize their battery life. Unfortunately, this life creates enormous problems for cellular towers. From Ars Technica, we find that:

… the iPhone uses more power saving features than previous smartphone designs. Most devices that use data do so in short bursts—a couple e-mails here, a tweet there, downloading a voicemail message, etc. Normally, devices that access the data network use an idling state that maintains the open data channel between the device and the network. However, to squeeze even more battery life from the iPhone, Apple configured the radio to simply drop the data connection as soon as any requested data is received. When the iPhone needs more data, it has to set up a new data connection.

The result is more efficient use of the battery, but it can cause problems with the signaling channels used to set up connections between a device and a cell node. Cell nodes use signaling channels to set up the data connection, as well as signaling phone calls, SMS messages, voicemails, and more. When enough iPhones are in a particular area, these signaling channels can become overloaded—there simply aren’t enough to handle all the data requests along with all the calls and messages.

In essence, the issue of congestion at cellular towards may not be a result of ‘data hogs’ but a consequence of how smart phones are being engineered; the backhaul networks are in fine shape, but the particular towers are being overwhelmed. (We might have a discussion/debate on the condition of middle-mile backhaul, but I’m less familiar with those stats so I’m leaving them out.) In the US, let’s not forget that areas with high penetration of smart phones like the Eastern and Western seaboard cities are also home to (a) expensive real-estate; (b) people who oppose the development of large cellular towers in their neighbourhoods (often out of fear of depreciating that expensive real-estate). The first condition means that there is an economic challenge to building towers, and the latter focuses on civil resistance to development that could better balance the ‘smart phone load’. In what I’ve read about deep packet inspection in the mobile market, DPI doesn’t alleviate this kind of problem – it doesn’t ‘correct’ of battery-saving engineering – but where the appliances can be installed under the guise of data hogs they could subsequently be used for (vendor stated) purposes of billing and service differentiation/provision.

At last year’s Canadian Telecom Summit, various vendors and sessions discussed the need to have better customer transparency to ‘improve the customer-service provider relationship’. In essence, they were really describing a desire to develop ‘insight’ into what customers do on mobile and wireline networks to better monetize those relationships, and in almost all cases the same vendors and speakers acknowledged the need to deal with the ‘creepy feeling’ that comes with using DPI and related surveillance architecture for pico-marketing purposes. Few had clear solutions to this, what is arguably the most significant PR issue that limits ISPs’ expansive uses of DPI for profit generation in the wireline and wireless environments.

It’s significant that the only really pressing statement about needing to efficiently manage bandwidth and spectrum came from RIM’s Mike Lazaridis. He was very concerned with actual mobile data usage, and at no point was traffic shaping or monitoring a ‘solution’: improved data compression techniques were the focus of his talk. Here, we had a smartphone vendor focusing on data efficiency itself instead of advocating for traffic shaping and analysis innovation. This strikes me as being very ‘forward thinking’, insofar as it tries to address the underlying issue of data growth instead of trying to find (what I see as) a bandage solution to limit and subsequently monetize this growth. Don’t stop people from using wireless, but learn how to transmit and receive it in a very efficient manner to deliver a solution that customers actually want!

However, it’s important that we not lose sight of the fact that DPI isn’t exclusively intended for traffic management but also to extend vision into the data stream – a data stream, I might add, that unless encrypted is almost entirely public facing. The drive to this transparency is confirmed by Allot Communications’ Cam Cullen, who last year noted that,

[t]he stats and visibility that DPI provides at the access and applications layer lets mobile operators build better service plans for congestion control and feed that data into mobile billing systems to support things like roaming and advice of charge.

To begin: data transparency can facilitate improved service offerings and let network providers more intelligently expand and develop their networks. You can learn what parties you might want to approach about setting up content delivery networks, what the trends in data usage are, and so forth. All of this can be invaluable in making the hard decisions of what to invest in, where, and why.

It is the potentialities of price, service, and customer discrimination that worry both PIAC and doctoral student Fenwick McKelvey. Having spoken with representatives of PIAC, the worry is that any such discrimination will generate inefficient economic situations for customers – they’ll pay higher costs, for less – and McKelvey similarly questions the fairness of any such differentiations. As I’ve understood McKelvey’s position, DPI facilitates an artificial differentiation based on economics rather than efficiency: you charge more for streaming video because it’s profitable, not because the economics of charging for streaming video improve the network situation.

I would suggest that it is a combination of DPI’s bad press, and the kinds of stories that would be generated if telecommunications providers stated they were using DPI for billing reasons, that billing isn’t a public-facing explanation ISPs present for introducing DPI into the mobile environment. I would, however, assert that billing and advertising are key motivating factors: the former allegation based on vendor statements and the latter based on less-public vendor statements and ISP discussions I’ve been privy to.

Even after all of this, there is a question of ‘so what?’

Where DPI and other traffic management solutions are publicly deployed to attend to a particular area of the telecommunications companies’ business (e.g. bandwidth management, customer billing, tiering service) it strikes me as dishonest to subtly extend them into other areas of the business without first, very publicly, alerting both customers and appropriate regulatory bodies. In the case of advertising, Canadian ISPs are expected to report to the OPC and CRTC if they ever decide to use DPI for advertising or discrete user-history analysis, but the same ISPs arguably have to meet a low bar to meet the OPC’s disclosure requirements concerning their actual uses of DPI.

Far more significantly, whereas the CRTC has maintained a hands-off approach to regulating wireless – and thus implicitly permitted discrimination in the wireless environment – that might be changing. In the forthcoming hearings in October and November entitled “Proceeding to review access to basic telecommunications services and other matters” we might see the beginning of an anti-discrimination regulatory framework that would address the economics of cellular service. The CRTC has been dropping ‘wireless’ into more regulatory hearing documents recently, which may suggest a transition away from their general forbearance approach and, if so, I expect that the work performed by PIAC and others will be leveraged to try and establish a policy of non-discrimination in the Canadian mobile market. Any such regulation will likely have a very real impact on the permissibility and conditions of deploying DPI appliances in mobile networks, as well as publicly lay bare whether any Canadian ISPs are interested in, or already moving to implement, mobile traffic management facilitated by DPI appliances. If last year was the year of wireline network management/neutrality in Canada, we might get lucky and see this one as the transition year that leads to a public discussion about Canadian telecommunications companies’ wireless network management/neutrality practices, with discrimination as the focal topic.

[1] For the past eight or nine months I’ve been preparing for and writing doctoral candidacy exams. The last exam was recently concluded, leaving me with the new title of ‘Doctoral Candidate’ and time to get back to what I love: network analysis, surveillance, and digital privacy!

One thought on “Deep Packet Inspection and Mobile Discrimination

  1. Good post Chris. On the wireless signalling channel overload issue, that is very similar to a SYN flood issue on Ethernet networks. The mechanism for establishing the data channel gets flooded, and consequently no data is ever transmitted. As you point out, it’s not a capacity/congestion issue in the traditional sense.


Comments are closed.