Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 14 Nov 2012 22:13:45 +0100
From:      Stefano Marinelli <postnet@dragas.dyndns.org>
To:        freebsd-net@freebsd.org
Subject:   "Weighted round robin" for LAGG - anyhow?
Message-ID:  <646662D0-E66C-4168-B351-482B06ACF26A@dragas.dyndns.org>

next in thread | raw e-mail | index | archive | help
Hello everybody,
I've been trying to do some experiments to improve my ADSL speed. The =
idea is to bond two ADSLs,  create two OpenVPN TAP channels connected to =
a remote (fast connected) server and doing a round-robind LAGG =
aggregation on both nodes. The remote end will NAT. The operation is =
successful but, as by design of the R-R configuration, I just get the =
the slowest link speed * 2. "loadbalance" mode is way slower.
The first ADSL is more or less 2.2 Mbit/sec, the second one about 1.4 =
MBit/sec so the gain is minimal. It's useful just for fault-tolerance.
I thought about doing  some sort of "weighted round-robin", giving a =
weight to the two tap interfaces. Something like "send 2 packets to the =
first one, than one to the second one".
I found some patches [1] for Linux (I'm compiling a kernel right now), =
but no information about FreeBSD. My tests show that FreeBSD has a =
better bonding throughput (in Linux, I just get a bit more than 2.3 =
Mbit/sec, on FreeBSD I can almost get the theorical 2.8Mbit/sec).
Did anyone ever tried something like that? I tried to have a look at the =
if_lagg source code, but I think it'd be quite difficult for me to do =
something like that, even trying to adapt those Linux patches.=20

[1] =
http://sourceforge.net/projects/bonding/forums/forum/77912/topic/2048022

Thank you,
Stefano=



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?646662D0-E66C-4168-B351-482B06ACF26A>