Book excerpt: Video multicast offers a scalable solution for large-scale television distribution. For commercial-grade video multicast services such as Internet Protocol Television (IPTV), it is often challenging to meet a set of conflicting design goals, including high video quality, low request response time and good system scalability. This dissertation addresses these issues by developing a decentralized peer-assistance architecture and integrating it with streaming-specific source and channel coding techniques to deliver multicast video in a timely, reliable and scalable way. Video quality is negatively affected by network packet loss. An existing framework to simultaneously address packet loss and request response delay is to combine multicast with a unicast service that provides both packet loss repair and fast stream startup functionalities. Typically, servers that provide the unicast service could become the system bottleneck. To mitigate this problem, we propose an alternative distributed solution to partially shift the burden of the unicast servers to receivers of the multicast. Using a Peer-Assisted Repair (PAR) protocol, we demonstrate that packet repairs can be delivered reliably on time using a combination of server-peer coordination and redundant repairs. PAR can be naturally extended to a Peer-Assisted Startup (PAS) protocol to facilitate fast stream startup, where a missing portion of the multicast stream is treated as a long erasure burst to be repaired. In these applications, we show that video transcoding can be seamlessly integrated to provide improved robustness to packet loss, reduced stream startup latency and improved system scalability. Erasure correcting coding plays an essential role in the peer-assistance architecture. To understand how to mitigate its delay into the stream startup process, we explore the design of erasure correcting codes that takes into account a decoding delay constraint. Modeling the peer-assisted unicast service as communication over a set of parallel links, we propose a practical construction of parallel-link block erasure correcting code, where each source packet is decoded at the sink on-the-fly with a strict decoding deadline. The proposed code handles two types of common errors in the upper layers of packet-switched networks -- bursty packet loss and link outages. We further show that the proposed code construction is delay-optimal among all codes that achieve the Singleton bound. The proposed decentralized architecture requires in-network video transcoding at the multicast peer receivers, where it is often desirable to use a scheme with low complexity, such as open-loop requantization-based transcoding. A known drawback of this scheme is the requantization distortion drift problem. We develop an analytical model to capture the distortion drift effect and show that this drawback can be largely mitigated through parameter optimization. We demonstrate its effectiveness in the transcoded-to-primary stream switching problem in fast stream startup.