Empirical Evaluation of DASH Streaming over Multipath TCP in a Mixed Traffic Environment.

Multipath TCP (MPTCP) is a transport layer protocol that enables multi-homed devices to simultaneously use more than one interface. As a result, MPTCP provides a higher bandwidth to the applications and also increases resilience to potential connection failures. These aspects are crucial for real time traffic, especially streaming applications. One of the open source streaming protocols which has gained popularity in recent years is Dynamic Adaptive Streaming over HTTP (DASH). Studies evaluating the performance of DASH streaming over MPTCP have consistently shown significant performance improvements in application QoE (Quality of Experience) under various scenarios such as packet loss, shared bottleneck, and variable path latencies. However, when MPTCP is deployed for streaming over the Internet, it has to compete with several other types of traffic. So far, no study has been conducted to evaluate the performance of DASH over MPTCP under varying traffic loads. Besides, all the existing studies have used the out-of-tree Linux implementation of MPTCP. In this work, we present a detailed evaluation of streaming DASH traffic over MPTCP and also compare its performance against streaming over TCP under mixed traffic environment. We leverage Linux network namespaces for creating the testbed and stream DASH traffic. Moreover, we analyze the impact of competing traffic on the throughput of MPTCP. Our work is also the first to use the state-of-the-art upstream implementation of MPTCP in Linux. We observed that while MPTCP offers certain benefits for streaming, its performance may not be significantly better than traditional TCP under varying traffic loads. Our research indicates that the effectiveness of MPTCP for DASH streaming largely depends on segment length. MPTCP tends to perform better within a specific range of segment lengths that are neither too small nor too large.