Interface Split Routing for Finer-Grained Traffic Engineering Abstract.pdf

Interface Split Routing for Finer-Grained Traffic Engineering Saqib Raza, Department of Computer Science University of California, Davis Chen-Nee Chuah Department of Computer and Electrical Engineering University of California, Davis Abstract Legacy IP routing restricts the efficacy of traffic engineering solutions. This re- striction stems from the constraint that traffic at a node must be uniformly split across all next-hop nodes corresponding to equal cost shortest paths to a destina- tion [1]. Proposals that alleviate this constraint either completely overhaul legacy IP routing e.g. [2–5], or introduce complex control and/or forwarding plane compo- nents [6–8]. This additional complexity departs from the elegant simplicity of legacy routing protocols where statically optimized [9–14] link weights embed all traffic engineering semantics. We present Interface Split Routing (ISR), which retains the basic forwarding and control mechanism of legacy IP routing. Furthermore, a set of link weights embed all traffic engineering semantics in ISR. However, ISR makes possible finer-grained traffic engineering by configuring independent sets of next-hops to a destination at each incoming interface. This lends itself well to modern router architectures where each incoming interface has its own forwarding table [15]. Consequently, at the aggregated node level, traffic to a particular destination may be non-uniformly distributed across next-hop nodes. Hence, ISR allows additional flexibility in routing traffic as compared to default IP routing while retaining its simplicity. We conduct simulation studies on representative ISP topologies to compare ISR with traditional link-weight optimized routing. ISR reduces the difference between optimal routing and weight-optimized routing by 50%. Key words: Intra-Domain Routing, Traffic Engineering, Autonomous System, Link-State Routing Protocol, OSPF, IS-IS Email addresses: sraza@ucdavis.edu (Saqib Raza), chuah@ucdavis.edu Preprint submit