If you're reading this, you're probably using a version of the transmission control protocol, or TCP, a system that regulates internet traffic to prevent congestion. It works, and it's getting better all the time. But it was a system made by puny humans - surely our machine-overlords can do better.
Yes, and possibly as much as two or three times better, say the MIT researchers behind Remy, a system that spits out congestion-stopping algorithms.
To use Remy, an Internet-goer plugs in answers to a few variables (How many people will use this connection? How much bandwidth will they need?) and what metric they want to use for measuring performance (Is throughput, the measure of how much data is going through, the most important? Or is it the delay, the measure of how long it takes that information to travel?).
The system then starts testing algorithms to determine which works best for your situation. Testing every possible algorithm would be impractical, so Remy prioritizes, searching for the smaller tweaks that will result in the largest jump in speed. (Even this "quicker" process takes four to 12 hours.)
The resulting rules that the system spits out are more complicated than in most TCPs, according to Remy's inventors: while TCP programs might operate based on a few rules, Remy works out algorithms with more than 150 if-x-then-y rules for operating. The simulations sound impressive: doubled throughput and two-thirds less delay on a computer connection, and a 20 to 30 percent increase in throughput for a cell network, with a 25 to 40 percent slower delay.
Indeed, where a typical TCP congestion-control algorithm might consist of a handful of rules — if the percentage of dropped packets crosses some threshold, cut the transmission rate in half — the algorithms that Remy produces can have more than 150 distinct rules.
“It doesn’t resemble anything in the 30-year history of TCP,” Winstein says. “Traditionally, TCP has relatively simple endpoint rules but complex behavior when you actually use it. With Remy, the opposite is true. We think that’s better, because computers are good at dealing with complexity. It’s the behavior you want to be simple.” Why the algorithms Remy produces work as well as they do is one of the topics the researchers hope to explore going forward.
In the meantime, however, there’s little arguing with the results. Balakrishnan and Winstein tested Remy’s algorithms on a simulation system called the ns-2, which is standard in the field.
In tests that simulated a high-speed, wired network with consistent transmission rates across physical links, Remy’s algorithms roughly doubled network throughput when compared to Compound TCP and TCP Cubic, while reducing delay by two-thirds. In another set of tests, which simulated Verizon’s cellular data network, the gains were smaller but still significant: a 20 to 30 percent improvement in throughput, and a 25 to 40 percent reduction in delay.
I read a little about this, best part of it is that it can be implemented fast and relatively cheaply. But don't expect your ISP bill to go down lol oh no, they'll just keep charging you the same and stuff more users on the network.