| In this thesis, we propose a novel active reliable multicast protocol with congestion control (ARMC). Our protocol is a complete one that includes both an error recovery scheme and a congestion control scheme. The ARMC uses softstate storage to facilitate fast error recovery and efficient congestion control. The error recovery scheme includes five mechanisms: a simple NAK suppression and aggregation mechanism, an efficient hierarchical RTT measurement mechanism, an effective local recovery and scoped retransmission mechanism, and a periodical ACK mechanism. In a radical departure from popular end-to-end congestion control, our congestion control scheme is a router-assisted window-based hierarchical scheme. With flexible use of the resources in active routers, our congestion control scheme can not only let multicast flows have an adjustable share of bandwidth of bottleneck links with TCP flows, but also let two different multicast flows have a different share of bandwidth of bottleneck links according to the resources they (are allowed to) use. In addition, our congestion control scheme can also achieve inter-receiver fairness [TM98], [TE99] by limiting the effect of congestion on a specific link to a small region. ARMC can also achieve scalability and responsiveness. We implement and evaluate our protocol in NS2 and also compare performance with ARM [LS98] and AER/NCA [SS00], both of which are representative active reliable multicast protocols. The initial performance evaluation indicates that our protocol can achieve considerable performance improvement with limited support from routers. Our work further confirms that active networks can benefit some applications and become a promising network computing platform in the future. |
