Omniscient: Towards realizing near real-time data center network traffic maps

Abstract

In order to make measurement-based placement, an optimiser must make informed decisions. Currently, it is difficult to assign routes or assign resource commitments to network paths in data centers, as applications do not declare what is carried within each flow. We propose to provide insight into the traffic that traverses each network link, realizing a near real-time map of a data center’s network traffic. We present Omniscient, a system that aims to increase the visibility into the data center network traffic by computing link utilization broken down by application instance using OpenFlow stats in an SDN-enabled data center. The goal of the system is to inform application instance placement and redundant network path assignment in order to improve application performance and resource utilization.