This post is our submitted response to NSF’s call for expressions of interest in the Future Internet Architectures summit, which i am attending this week.
What scientific contributions will you bring to the discussion about Future Internet architectures?
As scientists, we are compelled to explore how the peculiar structure relates to the function(s) of complex networks. Many complex networks in nature share the peculiar structural character of the Internet, but they also manifest phenomenal behavior: they efficiently route information without any observable routing protocol overhead. This achievement is currently beyond the reach of man-made networks. The Internet still uses a 30-year old routing architecture with fundamentally unscalable overhead requirements. Yet in those 30 years, the Internet’s inter-domain topology has evolved toward a structure for which nature has superior routing technology, if only we can figure out how to use it!
The prospect of zero-overhead routing is sufficiently attractive that in our previous NeTS-FIND project we developed a new theoretical framework to study it. In our framework, nodes in real networks exist in a separate but related “hidden metric space,” which guides routing without overhead or topology knowledge. We found strong evidence that not only do hidden metric spaces underlie real complex network topologies including the Internet (http://www.caida.org/publications/papers/2008/self_similarity/), but that a greedy routing mechanism applied to such topologies and underlying spaces yields a maximum percentage of paths that successfully reach their destinations. Remarkably, these successful paths almost always are shortest, regardless of the hidden space structure (http://www.caida.org/publications/papers/2009/navigating_ultrasmall/). This explanation for why (if not how) complex networks are naturally navigable had sufficiently high interdisciplinary impact for recent publication in Nature (http://www.caida.org/publications/papers/2009/navigability_complex_networks/)
We have also developed a model of Internet growth which provides strong evidence that preferential attachment is a driving force behind Internet evolution (http://www.caida.org/publications/papers/2009/AS_evolution/). The model yields AS-level topologies with links annotated by AS business relationships (customer-provider or peer-to-peer), and suggests that preferential attachment must be related to economic realities of ISP business decisions. To our knowledge, this is the first Internet evolution model that is realistic, parsimonious, analytically tractable, uses only measurable parameters, and “closes the loop.” The last feature means that having all model parameters measured from real Internet data, and substituted in analytic solutions, we can predict peculiar structural and dynamical properties of the Internet.
Finally CAIDA contributes an active measurement platform (Ark) as well as vital data resources to the creation of an underlying discipline that formalizes our observations and understanding of large-scale, complex networked systems such as the Internet. Ark directly addresses a short-term call in the
Network Science and Engineering Council’s recently published research agenda, namely to improve the quality of measurement-driven research in the computer networking community and a broader range of scientific disciplines. Ark provides an opportunity to test and validate hypotheses about how the current Internet operates. We are planning modules for integration with other data sources, as well as external validation of measurements and inferences against reported reality help balance the inevitable trade-off between fidelity and utility of network models.
Discuss how your research ideas might contribute to an overall network architecture, where the focus is on the system as a whole and on the interactions among the components.
We are using our current FIND funding to investigate exactly this question, including implementing and potentially deploying greedy routing over hidden metric spaces in an experimental control plane such as LISP (http://www.ietf.org/dyn/wg/charter/lisp-charter.html and http://tools.ietf.org/wg/lisp/). This implementation/deployment initiative will require a coordinated effort among different groups working on future Internet architectures.
Indeed, there are many practical technical details that still need to be worked out. Which components can we deploy incrementally? For example, we must change the semantics of IP packets to hold hidden space coordinates of the packet’s destination. Since we cannot touch end systems, we need address family gateways that translate between IP and hidden space headers, similar to the mapping function implemented as part of LISP. Based on our preliminary estimates, the IPv6 header provide enough bit space to hold hidden coordinates, so that LISP does appear quite close to what we need at the control plane. However, the data plane changes are more involved. We have begun discussions with router vendors regarding implementation constraints.
And then we still have policy, security, ownership, trust, and business models to worry about. But information dissemination (e.g., routing and forwarding) is the core function of any network. Our approach is to modernize how we understand and implement this primary function as well as the associated implications for realistic future network architectures.
What is your experience in working collaboratively in a multidisciplinary setting, across disciplines and areas of expertise as well as across academe and industry?
CAIDA has extensive experience participating in as well as coordinating and hosting interdisciplinary conversations, as illustrated by the listing of our ISMA workshop series titles (http://www.caida.org/workshops/). Many of our workshops have focused on interdisciplinary conversations explicitly structures to bridge gaps between and build connections across domains. In August 2008 we co-hosted a workshop on Networks and Navigation at the Santa Fe Institute, a unique institution dedicated to multidisciplinary collaborations on complex systems.
A strength of our recent work is the composition of researcher skills including first-hand knowledge of operational and engineering realities of the Internet, expertise in the theory and practice of Internet routing and data collection, skills in mathematical analysis and modeling of complex networks, the ability to provide realistic approximations to analytically intractable problems, experience with large-scale network simulation and emulation, and the interdisciplinary capability to broaden the impact of this project to other disciplines. The collection of results we have achieved so far demonstrate that an interdisciplinary research team can make rapid progress in formalizing our understanding of large-scale, complex networked systems.