[52757] in North American Network Operators' Group
RE: per-packet load sharing in cef or dcef
daemon@ATHENA.MIT.EDU (Charles D Hammonds)
Thu Oct 10 20:35:15 2002
From: "Charles D Hammonds" <hammonds@attens.com>
To: "Feger, James" <jfeger@feger.net>,
"Adam Atkinson" <Adam.Atkinson@damovo.com>
Cc: <nanog@merit.edu>
Date: Thu, 10 Oct 2002 17:27:24 -0700
In-Reply-To: <Pine.LNX.4.33.0210100816560.4492-100000@stuttgart.feger.net>
Errors-To: owner-nanog-outgoing@merit.edu
sounds to me like a lab scenario. and while it may not be a good idea in a
production network, it should still work. gonna try labbing it up as well.
what code versions you running, Adam?
Charles
-----Original Message-----
From: owner-nanog@merit.edu [mailto:owner-nanog@merit.edu]On Behalf Of
Feger, James
Sent: Thursday, October 10, 2002 6:18 AM
To: Adam Atkinson
Cc: nanog@merit.edu
Subject: Re: per-packet load sharing in cef or dcef
I am not sure what is happening for you, but as a general rule of thumb I
don't run per-packet load sharing on circuits that are not 'the same'.
The need for packet re-ordering goes way up when you are running
per-packet on two types of circuit. Are the circuits provisioned for the
same speed at least?
-jf
On Thu, 10 Oct 2002, Adam Atkinson wrote:
>
> per-packet load sharing in cef / dcef just doesn't seem to want to work.
>
> I have two 7500s joined by a frame relay link and a fast ethernet.
>
> Traffic is coming in to one of the 7500s via a third link, and goes
> to a loopback on the second 7500.
>
> I have cef turned on
>
> I have "ip load-sharing per-packet" on every interface
>
> The FR and FE have the same ospf cost
>
> The routing tables and "sh ip cef" agree that there are
> two equal paths to the loopback.
>
> But all the traffic goes over one link.
>
> I've been through Cisco troubleshooting documents etc.
> and as far as I can see, everything is perfect.
>
> Also, a deja search reveals plenty of other people with
> similar problems, and no solutions I can see. Are we all
> missing something obvious?
>
> --
> Adam Atkinson
>
