In what many are hailing as a stunning advancement in internet-related technology, researchers at MIT have developed a new system, dubbed “Remy”, which is able to automatically generate TCP congestion-control algorithms. In case you don’t already know, TCP (or Transmission Control Protocol) is one of the main, or core protocols which drive the internet as we know it. In essence, TCP exists within the transport layer and ensures that packets are error-checked and in good order. TCP congestion avoidance algorithms on the other hand, like “Remy”, are able to essentially speed up the functionality of the internet due to the fact that they actively seek to cut down on “congestion”.
Of course this isn’t the first instance of something like this being developed, there have been others like Compound TCP, NewReno, BIC and TCP Cubic, but they’re not really anything like Remy. In short, Remy is a machine learning system which experiments with possibilities in order to determine the most efficient means of avoiding traffic-related congestion. Think of it as a trial-and-error learning protocol which actually learns from its mistakes and takes not of what’s working at any given time. Unlike other systems which are more or less dependent on preconceived notions about a network, Remy only asks users to input assumptions related to the number of proposed users on a network, etc…
The reason that this is such a big breakthrough has to do with the inherent difficulty of selecting the appropriate algorithm for each particular situation given the amount of time allotted to do so. Naturally, computers are much better at identifying changes and choosing the appropriate course of action to deal with them, especially when networking protocols and data transfer are involved. Often times so-called small changes in the networking environment lead to a dramatic decrease in performance, what Remy is able to do is build solutions very quickly, looking for the most direct solution in each case. Remy’s algorithms are also packed with more “district rules” (as many as 150 vs. the dozen or so which other congestion-control algorithms might contain). In short, this means that Remy is able to process possibilities which people might not even be able to immediately consider.
Of course Remy has yet to be implemented on a wide scale as of yet, but the researchers are confident that this could lead to an overall increase in internet speed across the board (for everyone). In other words, downloads would be faster, pages would open up almost instantly, etc… Currently, Remy is being tested and examined in an effort to uncover just exactly how its algorithms function and how it might affect conventional systems when added to them.