Efficient Packet Sequencing for Dynamic Multicast Layers

Most multicast flows such as IPTV are streamed at a constant rate using UDP. To fairly share the bandwidth with TCP streams, several multicast congestion controls have been proposed. Highly scalable and most efficient ones are receiver-driven and use dynamic multicast groups, which means that each group begins to send at a high rate and slows down progressively. Each receiver joins groups to obtain a fair rate and thus gets a different amount of the data sent by the source. Thereby, despite promising results in terms of fairness, it is a challenging task to efficiently use these dynamic groups and to send useful data whatever the amount of data acquired by a receiver. This article presents a sequencer which maps out application data to dynamic groups in an optimal way. Multiple applications such as file transfer or video streaming, can use this sequencer, thanks to a simple API usable with any buffer containing the most important data first. This API insures that a receiver subscribed to N% of the source rate obtains the N% most important data of each buffer sent by the source. This solution is experimented on two test beds providing ideal or realistic network conditions. The first results prove the sequencer optimal performances, as well as the agreement between the received rate and the importance of the received data. However, the second scenario shows that even with this optimal sequencer, a lossy network may impact on the performances of applications unable to use data containing gaps. But, an extra evaluation shows that a simple FEC mechanism is enough to remove most of this impact. Finally, to show the relevance of this sequencer to the application layer, we propose a file transfer software using a FEC encoding. Results show the file transfer efficiency as a download with a single receiver generates only little more overhead than TCP. Moreover, download time is almost independent of the number of receivers, and is already faster than TCP with only 2 competing downloads.