Programming Languages for Software-Defined Networks

Programming Languages for Software-Defined Networks

Bachelor Thesis, Master Thesis

Software-Defined Networks (SDNs) provide a new way to configure computer networks. Special-purpose network devices with tightly coupled data and control planes are replaced by programmable switches managed by a logically centralized controller. The communication between the controller and these programmable switches is carried out using well-defined APIs (e.g., OpenFlow). Instead of configuring devices individually, network policies are implemented on top of the controller and then used by the controller to instruct individual network switches.However, SDN APIs like OpenFlow closely resemble the features provided by the hardware. OpenFlow uses a set of prioritized match-action rules as abstraction, which makes it difficult to write sophisticated network applications. For example, an application supporting multiple tasks at the same time must merge the switch-level rules required by each of the tasks into a single set of rules that can be installed on the switches. To overcome these limitations, different programming languages for SDNs have been proposed, that provide higher language abstractions on top of OpenFlow, including abstractions for querying the network state, basic service composition or language support for network verification.Developing new SDN language features requires a comparison with already existing languages, usually, in order to show that they increase the expressivity of the language while providing at least the same performance at runtime. However, currently it is quite cumbersome to compare different SDN languages, since they are implemented on top of different host languages, like Python, OCaml, Java and they usually provide only a small set of simple example application that are not directly comparable with each other.The goal of this thesis is to develop an extensible testbed for SDN applications that allows to compare SDN programming languages with respect to expressivity as well as performance and to provide a set of small- to medium-sized example applications that can be used for benchmarking and comparing the available language abstractions. Based on the results of the experiments, the next step will be to propose new language features that address the limitations of current solutions.

Publications

  • Nate Foster, Rob Harrison, Michael J. Freedman, Christopher Monsanto, Jennifer Rexford, Alec Story, and David Walker: Frenetic: a network programming language. ICFP '11
  • Christopher Monsanto, Joshua Reich, Nate Foster, Jennifer Rexford, and David Walker: Composing Software-Defined Networks. NSDI '13
  • Tim Nelson, Andrew D. Ferguson, Michael J. G. Scheer, and Shriram Krishnamurthi: Tierless Programming and Reasoning for Software-Defined Networks. NSDI '14