Multicast Techniques for Bandwidth-Demanding Applications in Overlay Networks

This dissertation, "Multicast Techniques for Bandwidth-demanding Applications in Overlay Networks" by Cheuk-man, Mark, Tsang, 曾卓敏, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of thesis entitled "Multicast Techniques for Bandwidth-demanding Applications in Overlay Networks" submitted by TSANG Cheuk Man Mark for the degree of Doctor of Philosophy at the University of Hong Kong in July 2008 Multicast is a communication primitive for data distribution from a single source to multiple receivers. The popular IP multicast protocol does not scale wellandishardtodeploy. Applicationlevelmulticast(ALM)hasthusemerged asanalternatesolutiontosupportmulticasting. InALM, themulticastsource and receivers act as overlay nodes on top of the underlying physical network. Various ALM solutions have been proposed. They aim at reducing end-to-end delay of small-sized data, and avoid duplicating packets at physical links to reduce network loads. They are suitable for low bandwidth multicast (LBM). These methods however cannot meet the demands of large-volume data dis- tribution, where high receivers' throughput are desired. In a high bandwidth multicast (HBM) session, the unicast flows are long-living; they may easily clog the physical links involved. In addition, data could be partitioned into segments; each receiver could aggregate disjoint data from multiple peers to improve its throughput. Hence, LBM and HBM solutions should be different in their design. Inourstudy, wesolvetheproblemofapplication-levelmulticastforbandwidth- demanding applications. It is to determine data routing paths among overlay nodes to complete multicasting on large volume of data. Two research issues areinvolved forthesolutiondesign: bandwidthcontentionalongphysical links reduces receivers' throughput; receivers could also fail to perform data aggre- gation effectively due to skewed distribution of disjoint data providers. The iproblem is challenging since bandwidth contention at physical links appears dynamically but is hard to predict without physical topological information. Complexity increases when the routing path of every data segment has to be determined to facilitate continuous data aggregation at receivers. We argue thatexisting solutions could failtoavoid bandwidth contention and theirdata distribution strategies could hinder efficient data aggregation. We identify several data transfer patterns which are the primary causes of physical link bandwidth contention. We thus proposed an HBM solution, NOAH, to avoid these problematic patterns. NOAH specially builds a tree of receivers and each aggregates data from other tree members. The tree bounds thetrafficcarriedbytheoutgoinglinkofanetworktoavoidcongestion. During data dissemination, close receivers in the tree coordinate to request data; this reduces duplicated packets along their shared incoming links. We use a data arrival time prediction mechanism and a congested flow detection scheme to determine which receiver could forward data timely without inducing conges- tion. Theoretically, outgoingtrafficofeachreceiver couldalsobebounded. To support efficient data aggregation at receivers, a fastest rarest first data selec- tion algorithm is proposed to decide which data segment should be requested. Different from traditional data selection schemes, our approach shortens the time to make data diverse among nodes; more nodes can have data available for sharing. We coded NOAH for simulation. Performance study confirms the better multicast throughput in NOAH than some existing solutions. This justifies the importance of avoiding th