Context

The last decade has witnessed a strong evolution of the Internet: from a hierarchical, relatively sparsely interconnected network to a flatter and much more densely inter-connected network in which hyper giant distribution networks (HGDNs, - e.g., Facebook, Google, Netflix) are responsible for a large portion of the world traffic, becoming the de-facto main actors of the modern Internet. The very same set of actors have fueled the move to very large data center networks ( DCNs), along with the evolution to cloud native networking.

Typically, to face the multiple challenges associated to content delivery and client satisfaction, HGDNs and DCNs deploy massive network infrastructures.

This project focuses on three particular challenges associated to those massive network infrastructures.

Challenges

Challenge 1: Response Time to Issues

First, the network infrastructure must be running all the time, even in the presence of (unavoidable) equipment failure, congestion, or change of traffic patterns. Said otherwise, it means that HGDNs and DCNs must carefully engineer their network infrastructure to be able to ensure that issues are responded to within seconds. Network monitoring and measurements are thus of the highest importance for HGDNs and DCNs, though the available tools and methods have not kept up with the pace of growth in speed and complexity.

Challenge 2: Customer Satisfaction

Customers want to enjoy their content whatever the context in which they access it: at home behind a DSL gateway, on a mobile device in public transportation, at home on multiple devices at the same time, etc. In addition, customers want to experiment their content with the highest possible quality and the lowest delay without interfering with the network. Consequently, HGDNs, DCNs, and classical Internet operators must carefully engineer their network to ensure the highest Quality-of-Experience (QoE) on the user side.

Challenge 3: Automated Configuration

As the network grows bigger and bigger and as the number of changes in the network are more frequent, the Command Line Interface (CLI) is no longer the norm for configuring networks. Indeed, networks must be automated: this is this DevOps time where the new mantra is: "if a feature can’t be automated, it doesn’t exist". Data model-driven management improved the configuration management, as more and more APIs are being standardized. Next to configuration, the networking industry has been adopting (model-driven) telemetry as a compulsory step to stream any monitoring information to a collector.

Solutions

In this project, we tackle those challenging by relying on in-Situ OAM (and additionnal developments towards a Cross-Layer Telemetry) and the Diagnostic Agent (and additional developments).