On Thu, 12 Apr 2018, Jonathan Morton wrote:

Yes, and that's a policy I can support (as far as it goes). I'd much rather have, say, 100ms of bloat than 45000ms (which I've previously encountered).

So after some more testing I discovered I kind of need:

class mikabr-LE
  random-detect 50 ms 1000 ms
  bandwidth percent 5

Because the "50ms" is calculated out of the "5%", which means when there is no other traffic, "50ms" means 2.5ms of buffer with 0% loss probability. Unfortunately it means 50ms of buffer when there is lots of other traffic (when it's only getting 5-8% of total capacity).

But I imagine bufferbloating LE should have no negative consequences as there should be no interactive traffic marked with LE, right?

Mikael Abrahamsson    email: swm...@swm.pp.se

aqm mailing list

Reply via email to