Program synthesis for networks

Abstract

Software is eating the world. But how will we write all the programs to control everything from sensors to data centers? Program synthesis provides an answer. It increases the productivity of programmers by enabling them to capture their insights in a variety of forms, not just in standard code. In this tutorial, we focus on some challenges in programming networks, and we show how program synthesis algorithms can help. Developing network programs is difficult, as networks are large distributed systems. In particular, implementing programs that update the configuration of a network in response to events is an intricate problem. First, even if initial and final configurations are correct, subtle bugs in update programs can lead to incorrect transient behaviors, including forwarding loops, black holes, and access control violations. Second, if the update program reacts to events occurring near simultaneously in different parts of the network, naive implementations can lead to causality violations and conflicts. We present scalable program synthesis algorithms that produce network programs that are both correct by construction and efficient.