| Abstract: |
The coordination and management of inter-domain routing in the Internet are crucial tasks that happen “behind the scenes”, while we use the Internet for checking our emails,
surfing the web, making Skype video calls, watching NetFlix episodes or engaging in social networking. The basic protocol that governs inter-domain routing, i.e., routing between the diverse
administrative network domains, is the Border Gateway Protocol (BGP). Researchers have pinpointed several issues with the design and operation of BGP over the years, such as slow convergence
times or lack of end-to-end service guarantees, and have proposed various solutions and improvements. Despite this large body of proposals, BGP is extremely resistant to change, albeit the only
core protocol that virtually every Internet Service Provider (ISP) uses today to route packets between domains. Therefore, we need to find ways to address the requirements of new Internet
services and evolve routing while maintaining compatibility with BGP. We thus propose a new architectural approach which can be incrementally deployed in today’s Internet and which can
gradually tackle BGP-related challenges while its adoption gains momentum. Within the scope of this work, we address the challenges related to suboptimal convergence behavior and lack of
inter-domain service guarantees. The approach is general enough to be used for optimizing further aspects of inter-domain routing beyond the scope of this work. As our basic approach, we make the
case for gradual routing centralization across domains based on Software Defined Networking (SDN) principles and outsourcing mechanisms. The envisioned setting is composed of contractors offering
Routing as a Service (RaaS), serving—contiguous or disjoint—clusters (i.e., groups) of Autonomous Systems (AS), which are their clients. In general, contractors may operate on
abstract views of their client networks. We show how the proposed architecture may be used as the basis for such improvements, by investigating two concrete and prominent use cases. As a first
use case, the RaaS contractors talk to external ASes via BGP, while providing inter-domain optimizations and new routing services within their respective client clusters via SDN mechanisms. These
optimizations, such as more stable routing and better convergence, may also benefit non-client ASes. In fact, we show that routing convergence improves w.r.t. time and load of routing updates, as
the sphere of influence of RaaS contractors expands over hybrid BGP-SDN multi-domain networks. We have thus developed an emulation framework to run hybrid routing experiments. As a second use
case, we investigate RaaS contractors which work independently from BGP by performing overlay stitching of partial paths crossing ISP domains. The goal of this process is to form end-to-end
routes with certain properties, such as guaranteed bandwidth, latency or availability. The stitching can be implemented, e.g., using well placed and connected programmable switching points at ISP
interconnects. The setup on which we simulate our proposal is the rich overlay fabric which ISPs and their peering interconnections compose, utilized in a novel way for the provisioning of
guaranteed end-to-end services across domains. This setup is aligned to the status quo in the ISP ecosystem, in terms of inter-domain topologies and peering relationships. In fact, our research
is based on Internet eXchange Points (IXPs); these are physical infrastructures at which multiple ISPs exchange data with each other. In order to investigate the potential of the aforementioned
peering fabric in terms of AS-level path diversity and associated metrics, we further develop a graph transformation algorithm. This algorithm can be used for the computation of min-cuts on
arbitrary network graphs governed by policies. One of the application contexts of this computation is an inter-domain setting governed, e.g., by valley-free routing policies. Metrics which rely
on policy-compliant min-cuts can be calculated centrally by a multi-domain RaaS contractor in order to evaluate routing availability, multipath throughput and reliability against link failures.
We investigate the benefits, implications and outlook of the general approach and its associated use cases, supporting the following hypothesis: Inter-domain routing centralization, performed in
a staged and consistent manner, can help to evolve and improve Internet routing on the AS level. In the context of the studied use cases, the routing algorithms, protocols and services are
evolving on top of the RaaS contractor’s platform. This platform is the logically centralized routing control plane which operates on an abstract global view of the (inter-)network. |
