MAMBO: membership-aware multicast with bushiness optimisation

Fan, L., Trinder, P.W. and Taylor, H. (2008) MAMBO: membership-aware multicast with bushiness optimisation. In: 2nd International Conference on Distributed Event-Based Systems (DEBS '08), Rome, Italy, 1-4 Jul 2008,

Abstract

Multicast is a key component of applications such as mul- timedia streaming, multi-player gaming and generic publish- subscribe and event noti¯cation services. IP multicast has a number of technological, practical, and business obstacles [3], and thus is not widely used. An alternative, Application- Level Multicast (ALM), has been proposed to support sim-ilar functionalities. A survey of existing ALM systems [5]indicates that many ALM systems are built on P2P overlayinfrastructures and construct multicast trees among appli-cation participants, e.g. Scribe [2] uses Pastry [6]. This article presents Membership-Aware Multicast (MAM) and MAM with Bushiness Optimisation (MAMBO), which reuse the communication structure of a tree-based ALM sys- tem to track group membership, i.e. to record when peers join or leave the group. MAM and MAMBO are imple- mented on Scribe with Pastry and a series of experimental results compare performance with a conventional service- provider architecture. We show that MAM and MAMBO quickly detect, and e®ectively manage, the arrival and de- parture of peers. In addition, MAMBO delivers a number of other advantages. The Bushiness Optimisation limits the overhead placed on any peer, hence preserving the scalabil- ity of the ALM system. The implementation as Scribe policy plug-ins illustrates that MAMBO can readily be overlapped on any tree-based ALM, and requires few changes to the underlying technology.

• Item Type: Conference Proceedings
• Status: Published
• Refereed: Yes
• Glasgow Author(s) Enlighten ID: Trinder, Professor Phil
• Authors: Fan, L., Trinder, P.W., and Taylor, H.
• College/School: College of Science and Engineering > School of Computing Science