NeTS: Small: Managing Network Function Virtualization under Uncertainty

Purdue University, West Lafayette IN

Investigators

Abstract

Data centers are increasingly virtualizing network functions that have been traditionally implemented as hardware appliances. Today's network functions (such as firewalls and multimedia support systems) are costly to procure, deploy, and change. With virtualization, these network functions can instead be implemented as software running on low cost commodity servers. Their placement, size and connectivity can be adjustable on-demand though management and orchestration systems. This project seeks to develop the theoretic foundations and practical implementations to achieve provably efficient, low-cost, and robust Network Function Virtualization (NFV) operation in dynamic and uncertain environments. The project will develop online algorithms for placing, scaling and migrating virtualized network functions that take uncertain future network traffic and compute workloads into consideration. The investigators will model the variability and uncertainty of the compute load and network traffic in a rigorous mathematical framework to develop control algorithms with provable guarantees in cost and performance. The three principal research activities are: 1) exploring online NFV orchestration and scaling under uncertainty; 2) managing end-to-end queueing performance under uncertainty and cross-function dependency; and 3) designing algorithms and analyzing key performance and complexity bottlenecks.