netdev
[Top] [All Lists]

Re: [PATCH 2/5] TCP BIC 1.1 support

To: Baruch Even <baruch@xxxxxxxxx>
Subject: Re: [PATCH 2/5] TCP BIC 1.1 support
From: Stephen Hemminger <shemminger@xxxxxxxx>
Date: Mon, 11 Apr 2005 14:35:59 -0700
Cc: Andrew Morton <akpm@xxxxxxxxx>, "David S. Miller" <davem@xxxxxxxxxxxxx>, netdev@xxxxxxxxxxx, Injong Rhee <rhee@xxxxxxxxxxxx>
In-reply-to: <423B9415.2010906@xxxxxxxxx>
Organization: Open Source Development Lab
References: <20050318162211.366ca490@xxxxxxxxxxxxxxxxx> <423B9415.2010906@xxxxxxxxx>
Sender: netdev-bounce@xxxxxxxxxxx
On Sat, 19 Mar 2005 02:53:09 +0000
Baruch Even <baruch@xxxxxxxxx> wrote:

> Stephen Hemminger wrote:
> > This patch adds TCP BIC back in as a pluggable TCP congestion mechanism.
> > This version is closer to the TCP BIC 1.1 released for Web100.
> > The changes from 2.6.11 are:
> >     * congestion window undo fix
> >     * delayed ack compensation
> 
> This can cause overshooting if the other side doesn't do delayed acking, 
> did anyone consider the ABC (Appropriate Byte Counting) patch from 
> Yee-Ting Li? 
> http://marc.theaimsgroup.com/?l=linux-netdev&m=110917262615630&w=2
> 
> FreeBSD has an option to disable delayed acking so it's not that hard to 
> work against a host with no delayed acks at all.

I tested with FreeBSD today and see no overshoot, but I do see the undershoot
(with Reno as well), because of the extra cwnd/2 that you already observed.

<Prev in Thread] Current Thread [Next in Thread>