Traffic Control in Datacenter Networks

Abstract

The modern digitized world is supported on the backbone of datacenters - Applications ranging from sending emails, shopping online, using social media and playing video games, to name but a few, are all massively dependant on datacenters for storing and processing their data. The size of the datacenters has only been growing to meet this demand, currently containing tens or hundreds of thousands of servers, with plans being made already for datacenters with over a million servers~\cite{massdcs}! The use cases vary quite a bit, from private servers built to run applications for a single company, such as Google or Meta, to servers operated by service providers that are able to rent storage and computation resources in their datacenters to customers. In all of these data centers, the massive amounts of computation power required to drive these systems results in interesting and complex challenges and distributed systems and resource management/allocation problems. More and more applications, in the modern day, deal with very large datasets from diverse sources; The increasing sizes of both datacenters and the applications that use them lead to several scalability issues that we detail below. This thesis addresses some of the issues that arise.