Video continues to dominate network traffic, yet operators today have poor visibility into the number, duration, and resolutions of the video streams traversing their domain. Current monitoring approaches are inaccurate, expensive, or unscalable, as they rely on statistical sampling, middle-box hardware, or packet inspection software. We present iTelescope, the first intelligent, inexpensive, and scalable softwarized network middle-box solution for identifying and classifying video flows in realtime. Our solution is novel in combining dynamic flow rules with telemetry and machine learning, and is built on commodity OpenFlow switches and open-source software. We develop a fully functional system, train it in the lab using multiple machine learning algorithms, and validate its performance to show over 95% accuracy in identifying and classifying video streams from many providers, including YouTube and Netflix. Lastly, we conduct tests to demonstrate its scalability to tens of thousands of concurrent streams, and deploy it live on a campus network serving several hundred real users. Our traffic monitoring system gives unprecedented fine-grained real-time visibility of video streaming performance to operators of enterprise and carrier networks at very low cost.

Video continues to dominate network traffic, yet operators today have poor visibility into the number, duration, and resolutions of the video streams traversing their domain. Current approaches are inaccurate, expensive, or unscalable, as they rely on statistical sampling, middle-box hardware, or packet inspection software. We present iTelescope, the first intelligent, inexpensive, and scalable SDN-based solution for identifying and classifying video flows in real-time. Our solution is novel in combining dynamic flow rules with telemetry and machine learning, and is built on commodity OpenFlow switches and open-source software. We develop a fully functional system, train it in the lab using multiple machine learning algorithms, and validate its performance to show over 95% accuracy in identifying and classifying video streams from many providers including Youtube and Netflix. Lastly, we conduct tests to demonstrate its scalability to tens of thousands of concurrent streams, and deploy it live on a campus network serving several hundred real users. Our system gives unprecedented fine-grained real-time visibility of video streaming performance to operators of enterprise and carrier networks at very low cost.

New interactive video applications are increasingly emerging over the Internet; these interactive applications are characterized by high bandwidth requirements that fluctuates depending on end-user actions (e.g. less bandwidth is usually needed for stationary scenes). More importantly, this interactive class of services also involves a requirement for high responsiveness (i.e. low latency) from the network, in order to respond in real-time to end-user actions. One emerging service of this nature is 360° video streaming; another example is cloud-based gaming services. In this paper, we focus specifically on JPIP (JPEG 2000 Interactive Protocol) applications that support remote interactive video browsing with dynamic pan and zoom capabilities, as a highly representative example of the interactive service class. Existing network communication services are mostly agnostic to latency implications, and hence are not well adapted to such interactive applications. Meanwhile, explicit resource reservation protocols have not been widely deployed, and do not consider the time-varying dependencies that naturally arise in interactive applications. In this work, we leverage software defined networking (SDN) principles to support a proposed “interactive service” class. The main contributions of this work are a network-exposed application programming interface (API) that provides visibility into the state of the network, an SDN-assisted congestion control algorithm that utilizes network state information to achieve the desired low latency and high bandwidth utilization requirements, and a fair resource assignment algorithm that shares available bandwidth among interactive and non-interactive traffic dynamically – all without a reservation protocol.

Internet Service Providers (ISPs) are struggling to cope with the growing volume of streaming video traffic in their network, and the problem will only exacerbate as Virtual Reality applications proliferate. To classify and manage bandwidth for video streams, current practise is to either sample traffic for offline analysis or deploy middle-boxes for in-line packet inspection - such solutions are inaccurate and/or expensive. In this paper we present Telescope, a low-cost system comprising a commodity SDN switch and a commodity server, to identify and profile individual video flows at line-rate. We develop an architecture that dynamically manages flow-table entries to classify video flows with minimal mirroring of packets, we prototype our solution using a Noviflow OpenFlow switch, coupled with the Bro packet inspection engine and our application on a Ryu controller, lastly, we validate our solution with real video streams in a campus WiFi network, and test its scaling to thousands of video flows using a hardware-based traffic generator. We believe our solution offers great potential for real-time video classification in an operational network at very low cost.