On 2005-05-31, Kentaro A. Kurahone <kurahone@sigusr1.org> wrote:
> Hello all,
>
>       I've updated/cleaned up the TCP Eifel code that I've had sitting arou=
> nd,
> and was wondering if I could get some feedback and more testing before I co=
> mmit.
>
> What it is/what it does:
>  * Eifel detection/response algorithms. (RFC3522/RFC4015)
>  * Used the TCP timestamp option to detect reordering.
>  * Restores congestion state in case of reordering induced window shrink.
>
> Who will benefit:
>  * GPRS users, or others on quirky networks (Modem mayby).
>   (http://www.cs.helsinki.fi/u/gurtov/papers/ew02.html cites 12% performance
>    benefit, though that's in simulation.)
>
> How to use it:
>  * Apply the patch, recompile, reboot.
>  * The net.inet.tcp.do_eifel sysctl will control it.

Why tcp.do_eifel and not just tcp.eifel ? I thought there was some kind of
"policy" in sysctl names as: net.inet.ipsec.ecn, net.inet.tcp.cwm,
net.inet.tcp.rfc1323, net.inet.tcp.newreno, etc.

-- 
Rui Paulo <rpaulo@netbsd-pt.org>