The idea was originally proposed by Tom Nielsen in issue #26. In summary:

The research discusses a user-level technique to control congestion over TCP (they call it paceline). It uses various techniques to prevent the socket’s internal buffer (and any proxy server buffers) from filling up, so that higher priority messages can sent sooner and data in the queue supports aborting. It also discusses a “quick failover” approach to having multiple sockets on hot standby. They claim that when a socket gets too much packet loss, it’s better to just close it and resend the data over an alternate one.

The ACM article is a little more readable than the paper, but it’s behind a pay wall.http://dl.acm.org/citation.cfm?id=1730836.1730858

I think this is a link to the original paper. http://www.cs.ubc.ca/~aerbad/paceline.pdf

We will come back to this in the next round of feature additions.