A Video storage management system for soccer analytics

Abstract

Video is dominating consumer internet traffic. Restless internet users expect smooth playback and low latency when watching video content and vendors risk losing customers if this cannot be provided. Distributed storage systems specialized for delivering video content and for handling the high traffic this lead to, have been developed over many years. This thesis look into building and deploying a distributed video storage to deliver video content to an enterprise sport analytics web interface using open source solutions. We look into properties such as horizontal scalability, fault tolerance, security and privacy to outline an architecture that can scale horizontally and deliver video content in an efficient way to a HTML5 web interface. Our conjecture is that developing such a distributed storage solution from scratch is not feasible, and that other, accessible, storage solutions should be evaluated. We ended up with deploying OpenStack Object Storage on our cluster, integrating it with the web front end for uploading, managing and accessing video content. Our experiments test and evaluate the performance and boundaries of our system. We also discuss elasticity problems related to sudden spikes in interest, and solutions in the context of privacy and economic issues. The final deployed video storage system has replaced an old centralized approach that previously has been used. It is currently in production delivering video content to a web interface that is used for soccer analytics by our sport partner Tromsø IL. A case study has been made to observe how they have been using our systems for the last year.