The increase in the number of mobile devices accessing the Internet creates new problems. A specific case is when a large number of wireless clients try to access the same video stream in the same hot spot, which is a common requirement in situations such as sports and cultural events. The video frames are provided to the clients via multicast communication, which is unreliable and does not confirm the packets' reception. In heavy loaded conditions, packet losses increase and the overall quality of the stream reception degrades with strong asymmetries among clients.
Several reliable multicast protocols exist to cope with this issue. Some of them, as ReAct [1], RALM [3] and RMDP [2] are based on Automatic Retransmission Request (ARQ). However they are mainly designed for multi-hop networks, minimizing the number of messages transmitted across networks, but not in a single-hop. They also lack support for Quality of Service (QoS) and Service Level Agreement (SLA) management.
In order to compensate potentially high and asymmetric packet losses in a multicast transmission, we propose using a feedback repeat request mechanism per client. Receiving feedback from the clients allows reporting missing packets enabling the server to retransmit them.
When many clients are involved, a large flow of feedback messages (acknowledgements) arises, increasing the processing effort in the server and the load in the wireless medium. An uncoordinated transmission of acknowledgements increases contention in protocol stack buffers and collisions in the medium, reducing network throughput and downgrading streaming QoS.
In order to mitigate this problem, a feedback management scheduler is integrated in the video streaming server, in charge of scheduling the feedback sent by the clients. Their feedback frequency is then individually adjusted according to their erasure level or predefined SLA. For example, consider two clients with different erasure levels (e1, e2). Based on the clients' feedback, the scheduler realizes that e1 > e2 and computes polling frequencies f1 > f2. Different frequencies correspond to different chances of packet recovery leading to QoS differentiation.
Therefore, the proposed protocol allows increasing the number of nodes and correcting asymmetric losses, thus avoiding streaming QoS degradation, while keeping the total bandwidth occupied by feedback messages within bounds. Preliminary results confirm the desired scalability with reliable streaming, based on an efficient bandwidth usage. In this demonstration we will show an experimental validation of the proposed protocol, establishing its effectiveness in mitigating the unreliability of multicast wireless transmissions and enforcing differentiated QoS levels per client.