Workshop on Internet Routing Evolution and Design (WIRED)

October 7-8, 2003
Timberline Lodge, Mount Hood, Oregon, USA

Position statement of

Aditya Akella


          On the Effects of the Wide-Spread Deployment of Route Control Products
          and Overlay Routing Services (Aditya Akella, CMU)
          Recent years have seen route control and overlay routing products that
          allow users and end-networks to select wide-area paths for their
          transfers in a more informed manner. For example, multihomed
          subscribers at the edge of the network are increasingly employing
          route control products (e.g., RouteScience's "Path
          Control"). Similarly, customers of Akamai's SureRoute service receive
          access to a large, diverse overlay network to route traffic on. The
          primary motivation for these products is to provide end-network-based
          mechanisms for optimizing wide-area performance and reliability. While
          the deployment of these products and services is not very widespread
          today, we expect it to grow rapidly over the coming years.
          At the same time, the deployment of such route control mechanisms has
          given rise to concerns about their impact on the general well-being
          (e.g. the stability of routing and network load) of the network. For
          this reason, the questions below are critical to our understanding of
          where the state-of-the-art in end-to-end routing and route selection
          lies today and where it is headed in the foreseeable future:
          - What is the impact of the deployment of route control mechanisms and
            services on the operation of ISP networks and on the efficient
            functioning of the Internet as a whole?
          - Would these products cause route or traffic instability in the
            Internet and if so, to what extent?
          - What new mechanisms do we need to put in place to counter the
            potential ill-effects?
          These questions can be addressed via a combination of measurement and
          analysis. The first step here is to accurately measure prevalent
          end-network practices for achieving intelligent route control and then
          build models for such end-network behavior.  It is also crucial to
          understand, in general, what the best end-network strategies are for
          improving performance and resilience. This may help influence (and
          possibly, model) future product design too.
          The next step is to study the impact that both limited and wide-spread
          deployment of route control products can have on network
          operation. Since these products are not overly popular amongst
          end-networks today, this question cannot be answered using traditional
          measurement-based approaches. However, modeling, simulation and
          analysis could give us the answers we are looking for. One useful tool
          is game theory. The interaction between various intelligent
          end-networks and the Internet can be modeled as a game in which the
          end-networks are selfish agents trying to individually maximize a
          local goal, such as observed performance. The models for end-network
          behavior constructed above could prove very useful in such an
          If the above analysis shows that deployment of route control does not
          impact the stability of routes in a negative manner, network traffic
          and the efficient operation of the network as a whole, then we need
          not be too concerned about the proliferation of route control
          products. If, on the other hand, the analysis shows that these
          products can have a negative impact on how well the network functions,
          then we may have to work on measures to counter the
          ill-effects. Stated otherwise, "aggressive" end-network behavior must
          be sufficiently penalized and thereby discouraged.
          One way to achieve the negative incentives described above is to
          design novel pricing schemes (which may involve rewriting SLAs) to
          ensure that end-networks offer somewhat fixed, predictable load to
          their provider networks. The SLAs could be coupled with policing
          schemes at the ingresses of ISP networks which could, for example,
          rate-limit traffic or drop packets to discourage a particular choice
          of routes made by the end-network. Such schemes could help strike the
          right balance between the end-networks' attempts to improve
          performance and resilience, and the carriers' goal of ensuring stable
          traffic and routes, by factoring in economic benefit as the key
          incentive for socially conformant behavior.
          To summarize, it is unclear yet whether widespread use of route
          control products will disturb the stability of the operation of the
          Internet. This issue should be further explored by first identifying
          the various ways in which end networks can impact stability, then
          understanding the extent of the ill-effects, and finally designing
          pricing-based mechanisms to contain the ill-effects.