Study On Routing And Congestion Control For Multicast With Network Coding

IP Multicast is a network technology that allows one or more sources to send packets simultaneously to multiple destinations. Multicast can effectively save the network bandwidth, and has been widely used, including instant messaging, online games, online video conferencing, multimedia distance education., etc. In recent years, network coding has become a hot research area in information theory and networking. Under the network coding paradigm, intermediate nodes between source and sink(s) encode incoming messages and forward the coded ones to their down-stream nodes, which is different from the traditional service of relaying and replicating data messages only. Multicasting with network coding can significantly improve the performance of multicast, including throughput, robustness, scalability, and so on. The main research contents of the dissertation are as follows:1) A multicast routing protocol for network codingThe previous works on multicast with network coding mainly focus on multicast with network coding in overlay network. Multiple overlay links often span the same physical link and share the bandwidth. On the other hand, the characteristics of underlying physical links and the multicast topology are usually not available to overlay nodes, which induce to difficulty in choosing coding schemes. Thus, multicast with network coding in overlay network can hardly reach theoretical multicast capacity. We propose a Network Coding based Multicast Routing Protocol(NCMRP) at IP layer. With NCMRP, the source broadcasts routing packets to find multiple disjoint paths from the source to each sink, and the multicast topology is constructed with the multiple paths. We implemented a prototype network system which multicasts video files to analyze the efficiency of NCMRP and network coding gain. With the experiments, we concluded that NCMRP achieves a much higher throughput and maintains reasonable bandwidth cost.2) A TCP-friendly congestion control scheme for multicast with network codingThe previous works on network coding are mainly under the assumption that link capacities are fixed and known. In practical networks, it is hard for routers to get available link capacities which dynamically change. So congestion control is needed to make full use of bandwidth while avoiding congestion. In addition, such a congestion control scheme should coexist friendly with TCP, the dominant transport protocol in the Internet. Existing congestion control approaches are not suitable for multicast with network coding because:(1) Network coding performs encoding at routers to mix data across flows;(2) Network-coding-based multicast is realized with a multipath topology, where a node may have multiple upstream nodes. We presented a TCP-friendly congestion control scheme for multicast with network coding. Along with congestion control, we realized multipath multicast with multiple multicast trees and propose a multirate multicast service for each multicast tree, so that the multicast flow from source to a receiver can be viewed as several unicast-like flows. We also proposed a window-based congestion control scheme to achieve TCP friendliness. Our congestion control scheme achieves larger throughput than previous schemes and is TCP-friendly.3) A deadline and size aware TCP scheme for datacenter networksLarge-scale, user-facing online services have deadlines to provide better user experience. Datacenters provide such online services by partitioning user requests amongst layered back-end servers within a data center and the results from the servers are then combined by an aggregator server before the final response is sent back to the user. As a result, any back-end network flow that does not complete by its deadline will not be included in the response and typically reduces the response quality, and also induces to the waste of computational and network resources.We propose DSTCP, a TCP-friendly deadline and size aware congestion control protocol for datacenter networks. DSTCP adjusts the congestion window size of a flow according to its deadline and its size, as well as the dynamic congestion degree of network. DSTCP requires no changes to the switch software and hardware and can be easily deployed. Coexisting with TCP, DSTCP greatly reduces the number of missed deadlines and also achieves a higher throughput, meanwhile maintaining a smaller queue.