Principal Investigator Hari Balakrishnan
Project Website http://web.mit.edu/remy/
Remy is a computer program that figures out how computers can best cooperate to share a network.
Remy creates end-to-end congestion-control algorithms that plug into the Transmission Control Protocol (TCP). These computer-generated algorithms can achieve higher performance and greater fairness than the most sophisticated human-designed schemes.
What makes Remy different?
Up to this point, human designers have created TCP's congestion-control algorithms by hand. There have been many such schemes over the last 25 years: TCP Tahoe, Reno, NewReno, Vegas, FAST, BIC, Cubic (Linux's current default), Compound (Windows's current default), etc. As the Internet has matured and evolved, crossing oceans and including wireless networks and ultrafast datacenters, so too have protocol designers added bells and whistles to our congestion-control algorithms.
But while it is easy to lay out the rules that govern a single endpoint running TCP Cubic, the emergent behavior of a flock of machines running Cubic is complicated and difficult to characterize. It can even be unstable. Congestion-control schemes each embody implicit assumptions and preconceptions about the network, but we don't quite know what they are. And the teleology of TCP -- the end goal that these algorithms try to achieve -- is unknown in general, when connections start and stop over time.
We propose to flip this around. With Remy, human protocol designers specify explicitly their prior knowledge and assumptions about the network and a goal to strive for. Remy then creates an algorithm to run at the endpoints — possibly a complex one, but with simple emergent behavior: achieving the goal as best as possible on networks described by the stated assumptions.