Paparao Palacharla

Abstract:
Current frameworks for network programming assume that the network comprises a set of homogenous devices that can be rapidly reconfigured in response to changing policies and network conditions. Unfortunately, these assumptions are incompatible with the realities of modern networks, which contain legacy devices that offer diverse functionality and can only be reconfigured slowly. Additionally, network service providers need to walk a fine line between providing flexibility to users, and maintaining the integrity and reliability of their core networks. These issues are particularly evident in optical networks, which are used by ISPs and WANs and provide high bandwidth at the cost of limited flexibility and long reconfiguration times. This paper presents a different approach to implementing high-level policies, by pushing functionality to the edge and using the core merely for transit. Building on the NetKAT framework and leveraging linear programming solvers, we develop techniques for analyzing and transforming policies into configurations that can be installed at the edge of the network. Furthermore, our approach can be extended to incorporate constraints that are crucial in the optical domain, such as path constraints. We develop a working implementation using off-the-shelf solvers and evaluate our approach on realistic optical topologies.