Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 12 Feb 2001 11:02:52 -0800 (PST)
From:      wpaul@FreeBSD.ORG (Bill Paul)
To:        bj@zuto.de
Cc:        ticso@cicely.de, net@FreeBSD.ORG
Subject:   Re: call for testers: port aggregation netgraph module (fwd)
Message-ID:  <20010212190252.2430F37B491@hub.freebsd.org>
In-Reply-To: <Pine.NEB.3.96L.1010211222117.30518f-100000@fledge.watson.org> from Robert Watson at "Feb 11, 2001 10:21:40 pm"

next in thread | previous in thread | raw e-mail | index | archive | help
> Date: Sat, 10 Feb 2001 15:29:17 +0100
> From: Rainer Clasen <bj@zuto.de>
> To: net@FreeBSD.ORG
> Cc: ticso@cicely.de
> Subject: Re: call for testers: port aggregation netgraph module
> 
> On Thu, Feb 08, 2001 at 01:25:09PM -0800, Bill Paul wrote:
> > http://www.freebsd.org/~wpaul/FEC/4.x/fec.tar.gz
> 
> I've tried this with 4.2-RELEASE on a single PIII-500 with 128 MB RAM and
> 2 fxp0 Interfaces. It was attached to a Nortel Baystack 450 running a 2
> port "Multinlink Trunk" Configuration.
> 
> It worked absolutely flawlessly. 

Ok...
 
> When downloading a single 60MB file to 3 clients by FTP, it achieved
> ~21 MB/sek output (netstat -I fec0 -w10). I had to use 3 clients due to
> lack of a second speedy client. 
> 
> I'm impressed. 

So am I.
 
> Am I right that a single transfer is always limited to the bandwidth of a
> single interface? The above mentioned Switch and Linux ditribute traffic
> in a round robbing manner. Both cooperate with Cisco's Etherchannel
> (tested against a Catalyst 2924). How about adding a round robbing
> distribution, too? This is epecially usefull for getting higher throughput
> when traffic mostly takes place between 2 hosts.

You are correct in your assumption. The port selection is done by XORing
the lower bits of the source and destination addresses in the packet
(either MAC address or IP address). So all packets transmitted to a given
host will always be sent over the same interface in the bundle (unless
it's down, in which case it'll use the next one over).

As for using a round-robin distribution... well, the idea behind
etherchannel is to distribute the traffic across multiple links while
preventing any out-of-order packet transmissions with minimal work.
Using a round-robin scheme would require some sort of mechanism to
deal with preserving packet ordering. Adding another port selection
scheme should be pretty simple as long as somebody can point me at
the right algorithm.

-Bill


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-net" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20010212190252.2430F37B491>