Improving multimedia streaming from the network's core

Abstract:

In 2015, NetFlix made the headlines with nearly 37% of Internet traffic being caused by their streaming applications. The trend towards similar applications is further accelerated by increasing number of mobile devices used to consume high resolution video and interact with low latency. Hence, new challenges arise that can no longer be solved by solely altering the network's edge. Software-Defined Networking (SDN) and its most popular incarnation OpenFlow [1] cause a trend towards intelligent devices at the network's core. The end-to-end (E2E) principle, as proposed by Saltzer et al. [2] in 1984 and further discussed by Moors [3] in 2002, is one of the paradigms to be reconsidered in this new world. Theory and practice show that segmenting multiple transmission domains can provide benefits for the overall network performance. Current approaches, as e.g. Split-TCP [4], require changes to the end-hosts and applications to explicitly use this protocol, which can be avoided when using SDN technologies. We propose the novel networking pattern Transparent Transmission Segmentation (TTS) that allows to separate certain domains and lead to better performance, especially in heterogeneous environments. With most of multimedia streaming traffic being transmitted using Dynamic Adaptive Streaming (DASH) [5], it is straightforward to consider TCP transmissions. We identify network functions of TCP that can perform better when segmentation is employed and give a differentiated view on this using simulations measuring transmission times. The evaluation shows that although we have performance degradations of maximally 1% for mean, median and standard deviation for certain scenarios, our approach can achieve more than 14% improvement on certain scenarios and 3 - 10% on most scenarios. This shows that TTS has to be applied thoughtfully, but can provide significant improvements when employed in the scenarios with high loss and jitter.