An RLNC transmission can recover from errors with neither sender nor receiver retaining and updating transmission-state information and requesting lost packets to be retransmitted. This is because RLNC can recreate any packet lost on the receiving side from a later sequenced packet. In over-simplified terms, each RLNC encoded packet sent is encoded using the immediately earlier sequenced packet and randomly generated coefficients, using a linear algebra function. The combined packet length is no longer than either of the two packets from which it is composed. When a packet is lost, the missing packet can be mathematically derived from a later-sequenced packet that includes earlier-sequenced packets and the coefficients used to encode the packet.
Since the RLNC encoding sender doesn’t need to listen for acknowledgements of successful transmission and perhaps retransmit, the sender can continuously transmit at near-wire speed optimized for latency and network throughput.
Update: After posting this I remembered I had read about an algorithm recreating earlier lost packets from future packets. So I clicked on the mit tag and on 10/25/2012 I posted this blurb: A Bandwidth Breakthrough
… The technology transforms the way packets of data are sent. Instead of sending packets, it sends algebraic equations that describe series of packets. So if a packet goes missing, instead of asking the network to resend it, the receiving device can solve for the missing one itself. …
That must mean they’re still working on it.