Adding Enhanced Services to the Internet

The 43rd edition of TPRC, the world’s premier conference on Internet policy, took place in Arlington last week, from Sept. 24-27th.  While this is mainly a policy conference that stresses law and economics, a few techies manage to find places on the program every year. I’m going to highlight the outstanding papers with substantial technical content over the course of the next year, both in the blog and in the podcasts. It’s fascinating to see technology informing Internet policy discourse, even if it’s a bit depressing to realize how rare this is.

For today’s post, let’s take a look at Adding Enhanced Services to the Internet: Lessons from History by kc claffy of CAIDA and David Clark of MIT. Clark is one of the authors of the famous – and famously misunderstood – paper End to End Arguments in System Design that lays out a principle of distributed systems building that has become foundational to Internet architecture and policy. He was also the Internet’s chief architect for nearly ten years, and is a member of the Internet Hall of Fame.  Klaffy is a genius at Internet measurement who founded and works at CAIDA, the Center for Applied Internet Data Analysis.

The paper makes five key points:

  1. “Quality of Experience” across all application classes is important to Internet users but is hard to measure.
  2. Service differentiation by application (or class of service) has been an Internet engineering goal from the beginning.
  3. The mechanisms for providing service differentiation have been well developed since the 1990s.
  4. Because of regulatory constraints and (to a lesser extent) business complexities, Quality of Experience differentiation is confined to private networks today.
  5. If differentiation is not allowed on the Internet as a whole, applications that need more stringent delivery controls than standard web surfing and/or video streaming will be relegated to private IP networks segregated from the Internet.

Here are quotes from the paper that address these points.

QoE is hard to measure and predict:

As a practical matter, the linkage between QoS and QoE is sometimes obscure; many technical quirks in the system can produce observable impairment in QoE. As well, our understanding of QoE and how to measure it is not as well developed as our understanding of the technical dimensions of QoS. So the argument that deployment of enhanced services is justified depends on a line of reasoning that is subject to argument–is the proposed differentiation actually going to mitigate the QoE impairment?

Service differentiation has been part of the Internet from the beginning:

The idea of enhanced or differentiated services has been a part of the Internet’s design from the beginning. The Internet Protocol (first defined by RFC 760 in 1980) had as a part of its header an 8-bit field called Type of Service, or ToS. Box 1 describes how the designers conceived of this ToS field at the time. This field enabled specification of service quality using five parameters, some of which were supported by some networks at the time: precedence, stream vs. datagram, reliability, speed vs. reliability, and speed. The specific precedence terminology mirrored then current controls in military command and control networks, and allowed the sender to specify the importance of the message.

The mechanisms for Quality of Service have been well-developed since the 1990s:

The intserv standards, which realize the guaranteed service, are specified in RFC 1633 (“Integrated Services in the Internet Architecture: an Overview” (Mo et al., 1994)). The core standard (RFC 2211) (“Specification of the Controlled-Load Network Element Service” (John Wroclawski, 1997)) described a complex set of mechanisms, including a flow setup phase where the sending machine specifies key parameters of the anticipated flow, and the state setup protocol RSVP (RFC2205) (Braden, Zhang, Berson, Herzog, & Jamin, 1997).

Regulatory and business constraints confine QoS to private networks (where it works very well):

However, at least this year in the U.S., the obstacle to enhanced QoS in this specific circumstance would be regulatory, because the FCC has ruled out the possibility for the content provider, as a customer of the access ISP, to pay for any service enhancement to the end user (Federal Communications Commission, 2015). The FCC termed this paid priority, and concluded that it violates the neutrality that the ISPs are expected to provide, both with respect to end users and content providers (what the FCC calls edge providers). The FCC has stated (Federal Communications Commission, 2015, paragraph 221) that service differentiation is more likely to be acceptable if it is application-agnostic, or if it is under the control of the user, thus presumptively not harming the user or his data flow. Unfortunately, as described in Section 4.3, there is currently no framework within which a user (or application code acting on the user’s behalf) could take steps to invoke or control the use of service discrimination.

The Internet can become fragmented because of diverse application needs:

Any theory of regulation that tries to limit activities of the platform owner must consider the degrees of freedom that the platform owner has, i.e., unintended side-effects. These alternative IP-based platforms greatly increase those degrees of freedom. Banning discriminatory traffic treatment on the Internet may prevent certain harms, but may also prevent the public Internet from competing with alternative private IP platforms with superior QoE, an undesirable outcome. A structured focus on potential harms and benefits of specific discrimination and pricing behavior of ISPs can help frame a debate about how to maintain the Internet as a vigorous platform that can compete with alternative private platforms. Specifically, allowing ISPs to sell QoS enhancements that lead to improved QoE on the public Internet may reduce the drive to offer specialized services, preserving the Internet as the unified service platform. It is worth exploring possibilities to accommodate such enhancements using monitoring to detect potential collateral impairments.

Conclusion:

The co-evolution of regulatory, legal, business, and technological capabilities, all at different paces, is tightly coupled in the case of enhanced services – a quintessential interdisciplinary challenge. While barriers to the deployment of scalable interprovider QoS on today’s Internet may be insurmountable, the issues and challenges remain. If any Internet of the future is to be the platform for a full range of useful applications, either the basic transport service must be so improved that no support for differentiated services is necessary, or it will be necessary to overcome these challenges. For this reason, it is worth developing a systematic understanding to the challenge of enhanced services, starting with cataloging their successes and failures over the history of the Internet as carefully as possible.

Are you eager for a podcast with the authors of this paper? I know I am.