Re: [patch 3/3] tcp: remove experimental variants from default list
From: David Miller <davem@davemloft.net>
Date: 2007-02-12 20:34:44
From: Stephen Hemminger <redacted> Date: Mon, 12 Feb 2007 12:26:34 -0800
No magic paper. But the impression from multiple talks is that Cubic is still doing fine. Also for non high speed flows, it really doesn't matter because all the loss based congestion controls behave the same.
This last point is important. Either it's high BDP and it uses the new stuff (and we WANT it to) or it's not high BDP and it USES RENO! That's why Baruch's email is totally illogical. If people don't have high BDP, these algorithms such as CUBIC and BIC change nothing compared to Reno, they purposely behave precisely the same.