This paper addresses the problem of streaming packetized media over a lossy
packet network, in a rate-distortion optimized way, where the rate-distortion
optimization is performed at the receiver. We show how the receiver should
compute which packets, if any, to request from the sender, in order to meet
a constraint on the transmission rate from the sender to the receiver while
minimizing the average end-to-end distortion. Experimental results show that
the receiver-driven system has performance within 1 dB of a sender-driven
system, while reducing the computational burden at the sender.