On 01/14/16 at 12:48P, Randall Stewart via freebsd-transport wrote: > > As per our discussion on transports? > > Here is a interesting CC algorithm that has been recently pointed out to me? > > Not sure its right (it may be ok .. or not) .. its something to think about > though... > > http://arxiv.org/pdf/1409.7092v3.pdf
Thanks Randall. Yeah, I now remember looking at it briefly. More info: http://modong.github.io/pcc-page/ The whole idea is indeed...interesting. In simple terms instead of reducing cwnd on (perceived) loss or congestion event, idea is to try and experiment with different sending rates and compare the results (based on what you are looking for in a connection) to see which of the approached helped you achieve best outcome. I've yet to go through all the details in the paper. But their claims seem very promising based on whatever testing they've done using the prototype implementation in userspace on top of UDP. Not sure if we can easily port the implementation into our modular CC framework. Cheers, Hiren
pgp4XkVJKmaaS.pgp
Description: PGP signature
