A group of MIT researchers has unveiled a machine learning approach to TCP congestion which could form the foundation of the next round of improvements to the venerable protocol's performance. Dubbed “Remy”, their TCP control software is based on the idea that even sophisticated modern congestion control algorithms (like …
So make the network self learning with a better model?
Why do I keep thinking "positive feedback" oscillation?
Interesting idea but one among many I suspect.
A learning network?
That can reprogram itself? I can't see that going wrong at all....
Re: A learning network?
Well done, you are able to paraphrase the sub-heading. Have a gold star.
It wont be running "out there" for a while
From their pub:
>>We have implemented Remy. Running on a 48-core server at MIT, Remy generally takes a few hours of wall-clock time (one or two CPU-weeks) to generate congestion-control algorithms ofﬂine that work on a wide range of network conditions.
It's a nice idea but I don't see it being used widely given that dynamic seems to imply a few hours on a 48 core box. It's not going to be dropping in on my PC any day soon let alone my phone.
sq_codel works well for me so far and needs no configuration and it has the added benefit that there is a fairly simple algorithm behind it that I can understand.