Network Mischief Managed: Automating perfSONAR Deployment, Management, and Measurement across a Global Campus
Time 10/16/18 08:50AM-09:10AM
As campus networking environments grow ever more complex, granular and accurate measurements have become crucial to discovering and debugging performance issues. Tools for obtaining and recording such measurements are the stock-in-trade of perfSONAR, but scattering perfSONAR nodes throughout the network can result in administrative nightmares - particularly when short on personnel.
We'll demonstrate how Ansible was used to automate the deployment and configuration of perfSONAR nodes across Duke's network, both in Durham and in Kunshan - including integration into MaDDash dashboards and Nagios alerting infrastructure. We'll delve into the horrors we found waiting in our network's "Chamber of Secrets" (hardware misconfigurations in our VMWare infrastructure, cross-VRF and VPN-tunnel performance anomalies), that were revealed by our currently deployed farm of perfSONAR nodes. We'll also discuss the challenges that we faced, while engineering this solution - and potential directions for the development of the perfSONAR suite that would help make wrangling an army of nodes easy, even for a sorcerer's apprentice.
Primary track Advanced Networking