On-Demand Layer Addition (ODL) : Making Multi-Layer Multicast Transmissions Cheaper
Résumé
The context of this work is the multicast transmission of data using multiple multicast groups, one per layer of information.One reason to use multiple groups is to address the potential heterogeneity of receivers and avoid that high-end receivers suffer from areduced reception rate because of the presence of low-end receivers.But there is a risk that some of the groups are not used.Sending data to a group with no receiver has a cost that is often underestimated: information is kept inrouters, periodic tree management traffic is required with both dense and sparse mode protocols, packets are sent on the LAN even if they are dropped by the first-hop router, IGMPv1 has limitations, etc. As IP multicast becomes widely deployed and used, these issues may well compromise its scalability. In this paper we introduce a new protocol, ODL (On Demand Layer Addition), which \emphenable- s a source to use only the layers (i.e. groups) that are actually required by the current set of receivers. We describe its behavior when used with several kinds of packet scheduling schemes (cumulative or not) and different scenarios (one-to-many versus many-to-many). We also show that this protocol can lead to significant savings when used with router filtering techniques, even if a single multicast group is used in that case. We have implemented the ODL protocol, integrated it in a multicast library and we report several experiments that assess its benefits.