Edinburgh Napier University

  Efficient application-level multicast (ALM) is an important research area supporting applications such as P2P file sharing, multi-player online games, and multimedia streaming. Although some applications only require a best-effort service, others have higher reliability requirements.

We propose a new reliable ALM mechanism, MAMBO, that reuses the ALM communication structure to monitor and organise its members. We show that a direct overlapping of multicast and supervision trees, MAM, can deliver reliable multicast, but may induce heavy workloads on some nodes, so a Bushiness Optimisation is introduced in MAMBO to limit maximum workload. Both MAM and MAMBO have been implemented as policy plug-ins for the Scribe ALM on the FreePastry P2P overlay. We show that both MAM and MAMBO effectively implement reliable application-level multicast except when parent and child nodes fail simultaneously.

We demonstrate the following advantages of MAMBO. Both MAM and MAMBO are more scalable than a conventional service-provider architecture, and MAMBO is more scalable than MAM. Both MAM and MAMBO reduce communication overheads compared with a service-provider architecture for applications with moderate levels of talkativeness. Because MAMBO reuses the ALM infrastructure, it induces minimal overheads and mininmal technology changes, e.g. it is implemented as Scribe policy plug-ins.