Adaptive media streaming over IP multicast (mABR)
Arsham Farshad, Mike Nilsson, Steve Appleby · 2022
Increasingly, major live events are being delivered using HTTP Adaptive Streaming (HAS), which, being a unicast technology, causes such events to generate huge traffic demand peaks on broadband networks which in turn drive significant investment in the network to increase capacity. Multicast ABR (mABR) is a streaming technology that inserts multicast into the path of a unicast HAS stream, allowing the traffic peaks to be reduced while requiring little or no change to the client players. But the unmodified clients continue to behave as HAS clients, asynchronously requesting content segments, adapting the quality of content requested according to perceived network conditions, totally unaware of the use of multicast to deliver synchronously one or more of the available encoded representations of the content to a local proxy. This creates inefficiencies and challenges in system design. Segments delivered synchronously must be cached locally ready for asynchronous requests from the client. Adaptation by the client will cause unwanted segments to be received by multicast, and alternative representations to be requested by unicast. We exploit CDN log data for BT Sport channels delivered by HAS to explore these inefficiencies and challenges. We show that unicast traffic in the core network could be reduced by more than 75% by delivering a single HAS representation by multicast to each client proxy, with larger savings for HD than UHD streams. We show that a cache at the client proxy capable of storing seven segments of 6s duration would be sufficient to be able to satisfy the majority of asynchronous requests from clients. Finally, we evaluate the performance of four policies for a client proxy to join and leave multicast groups in terms of the competing metrics of savings in unicast traffic in the core network and the quantity of data delivered over the access network.