I've been looking at this for the last several days (I'm giving a talk on it tomorrow.). If you're at all interested in computer networking or network-related software development, you might find this interesting.

http://pdos.csail.mit.edu/uia/sst/

The idea and prototype, developed by Brian Ford (MIT), in pretty interesting. In a nutshell, he talks about transport-layer hierarchical streams. A root stream (channel) used by an application can spawn child streams (similar idea to unix fork). That in itself it not that interesting, but when you consider that each child stream (aside from the root)...

1. is lightweight in that it does not incur 3-way handshake delays or TIME-WAIT delays

2. has its own (independent) flow-control context
(i.e. each stream can transmit at different rates and a blocking stream does not affect other streams in the channel.)

3. Can be a reliable byte stream or datagram model
(If datagram is chosen, but a datagram is too large ("large datagram problem") transparently/automatically falls back to reliable byte stream.)

...the coolness factor increases. Also, all streams in a channel share a congestion-control context, which helps it to "play nice" with other protocols.

Intersting read.

Wow I'm gonna keep looking at this, maybe a lag free streaming solution awaits!

Well, not necessarily. The biggest point of the paper is that applications can mix both a reliable byte streams approach (like TCP) and a datagram approach (like UDP), but without many of the inherent problems of each (e.g. The per stream costs of TCP and the datagram size limitations of UDP).

That is really interesting. Thanks for sharing

Sure thing. Found it in the ACM SIGCOMM publication. Plenty of good papers in there.

Just found the Brian's Structured Streams presentation he gave at Google (youtube).