Traditionally, a large number of dedicated media servers have been deployed to serve a large population of viewers for a single streaming event. However, maintaining media servers is not only costly but also usually requires over-provisioning due to the difficulty of predicting the peak size of an audience. Peer-to-Peer (P2P) streaming is a new approach to overcome these difficulties inherent in server-based streaming. We have developed the Stanford Peer-to-Peer Multicast (SPPM) protocol for live multicast streaming. SPPM constructs multiple multicast trees to push media streams to the population of peers, thereby achieving low end-to-end transmission delay. The degradation of video quality due to peer churn and packet loss in the network is reduced by video-aware packet scheduling and retransmission. In this paper, we present lessons we acquired from the deployment of a commercial variant of SPPM for a large-scale streaming event which attracted more than 33,000 viewers. We collected server logs and analyzed user statistics as well as the system performance. The results show that our system can achieve low end-to-end delay of only a few seconds with an average packet loss ratio of around 1%. We also found that improving peer-to-peer connectivity can substantially enhance the aggregate uplink capacity of P2P systems.