PhD Defence • Systems and Networking — Scalable and Reliable Middlebox Deployment

Please note: This PhD defence will be given online.

Milad Ghaznavi, PhD candidate
David R. Cheriton School of Computer Science

Middleboxes are pervasive in modern computer networks providing functionalities beyond mere packet forwarding. To name a few, load balancers improve network performance, intrusion detection systems protect networks from threats, and network address translators mitigate the IP address space scarce using address translation. Despite their benefits, middleboxes come with several challenges; their deployments are expensive, complicated to scale, and prone to failures.

The goal of this thesis is to devise middlebox deployment solutions that are cost effective, scalable, and fault tolerant. The thesis includes three main contributions. First, we discuss distributed service function chaining where multiple instances of a middlebox are deployed on different physical servers to optimize resource usage. Second, we introduce Constellation, a geo-distributed middlebox framework that enables a middlebox application to operate with high performance across wide area networks. Finally, we present a fault tolerant service function chaining system that provides a reliable deployment solution for a chain of middleboxes.