[GTER] CAKE e ack-filtering

Bruno George Moraes brunogm0 at gmail.com
Tue Oct 9 21:43:25 -03 2018


Pra quem nao ouviu falar dos beneficios extras de acabar com bufferbloat e
outras inovacoes do mesmo grupo.

https://www.bufferbloat.net/projects/codel/wiki/Cake/
Cake - Common Applications Kept Enhanced: Faz muita coisa(integral
shaper +  framing overhead compensation DSL ATM PPPoE + correct diffserv ).


http://blog.cerowrt.org/post/ack_filtering/

Extreme emulated test link 1Gbit/20Mbit,  with 100ms round trip delay.
Using  FLENT tool "rrul_be test", which starts 4 upload and 4 download
streams simultaneously:


Wow. At T+10 we get twice the upload throughput of the non-ack-filtered
case, and Download throughput is 20-60% better.

At T+30 upload throughput falls to only 50% more than before, (I don’t know
why, I think it might be due to cake’s “sparse flow optimization” on the
now thinned ack flows not queuing more acks to be filtered) - but at T+30
we are getting twice the download throughput than the unfiltered stream!
791Mbit vs 410Mbits!
Note: benefits to doing ack filtering only start to appear at poor
upload/download ratios. There might be some benefits also with 10GigE and
higher as those have trouble getting high pps for small packets.



More information about the gter mailing list