Janus - A Software-Defined Networking MPEG-DASH Video Streaming Load Balancer Edenilson Jˆ onatas dos Passos Graduate Program in Apllied Computing (PPGCA) Department of Computer Science (DCC) Santa Catarina State University (UDESC) Joinville, Brazil edenilson.passos@yahoo.com Adriano Fiorese Graduate Program in Apllied Computing (PPGCA) Department of Computer Science (DCC) Santa Catarina State University (UDESC) Joinville, Brazil adriano.fiorese@udesc.br Abstract—Recently, with popularisation of video streaming service, new video distribution technologies have been created. Currently, one of the most promising ones is the Moving Picture Expert Group Dynamic Adaptive Streaming over HTTP or MPEG-DASH. Even so, with the limitation of the TCP/IP network structure, the end user Quality of Experience (QoE) may be affected. One issue that can affect user QoE is the workload of content distribution servers. Thus, the unbalancing of server’s workload comprising user’s attendance can lead to a content server provider non optimised choice. This work presents two load-balancing solutions between MPEG-DASH video servers based on Software-Defined Networks, using as a balancing workload metric the throughput of the content server as well as the CPU load. Index Terms—Load balancing, MPEG-DASH, SDN, Flows manipulation I. I NTRODUCTION The use of the Internet on video transmission plays a relevant role in the business models of the current content providers. This is beneficial to users as it provides great control over when and where content can be viewed. A survey conducted by The Nielsen Company found that out of a total of 30,000 participants from 61 countries, 65% say they consume some Video On Demand content [1]. In addition, according to [2], by the year 2022, video traffic on the Internet is expected to reach 82% of the total. With the popularisation of streaming services, and conse- quently the overloading of the links as well as the restrictions imposed on this type of traffic by ISPs, companies began to develop adaptive video formats and content using the HTTP protocol as delivery method. Thus, the standard ISO/IEC 23009-1 or Moving Picture Expert Group Dynamic Adaptive Streaming over HTTP (MPEG-DASH) [3] appeared. A Content Delivery Network (CDN) is responsible for storing content that customers want as well as routing the requests to the closest server to the client. In this way, a CDN is responsible for load balancing between content servers [4], [5]. In these cases, load balancing is needed since, according to [6], server congestion is the factor that most affects the Quality of Experience (QoE) of the user because it is perceived as interruption and noise in the content’s image and audio. A contemporary approach to load balancing in CDNs for video content delivery is based on serving multiple client requests across multiple servers distributed in the CDN. To do this, the assignment of which server will attend to which client is done by means of the custom deployment of Domain Name Server (DNS) servers by the CDN provider. Thus, the DNS entries are adapted according to the adopted load balancing policy, in order to change the addresses of the servers that will serve the clients, hence performing the load balancing between the content servers. However, large-volume and long- term flows, such as video-on-demand (VoD) traffic, are hardly manageable using this DNS approach. The introduction of Software-Define Networking (SDN) is promising to solve this problem. Particularly, SDN instantiations, such as the one using OpenFlow protocol, allow the transparent redirection of flows in the network without the need for manipulation of infrastructure services [7]. Therefore, along these lines, the present article deals with the proposition of an architecture for load balancing in the MPEG-DASH video distribution network with the aid of SDN to perform the manipulation of necessary flows by means of the protocol Openflow version 1.3 and the Ryu controller. The balancing performed is intended to alleviate the workload of the content source servers and uses a combination of server throughput and CPU utilisation as balancing metric. This paper is organised as follows. Section 2 discusses related work. Section 3 architecture and operation modes of the proposed solution are detailed. Section 4 presents evaluation of the proposal. Finally, Section 5 presents final considerations comprising this work. II. RELATED WORK In [8] an adapted Dijkstra algorithm is presented for load balancing on a CDN using SDN technology. According to the justification used, algorithms that use random selection approaches or the Round Robin one, when selecting a route to the content server, can select the longest route (hops re- lated) and/or with congestion, dramatically decreasing service performance. With the proposed algorithm, the intention is to select the shortest path and with the least congestion. 978-3-903176-24-9 © 2019 IFIP