[nsp] Load sharing over unequal bandwidth lines using CEF

[Volodymyr Yakovenko]

On Mon, Jun 02, 2003 at 08:06:39AM -0400, Earl Zmijewski wrote:
>Hello.  I have what I thought would be a simple question, but can find
>no reference to a solution.  Consider a single router with 4
>equal-weight default routes out 4 interfaces.  The goal is to take
>advantage of the bandwidth across all interfaces, but they vary in
>bandwidth.  Using CEF will apparently randomly distribute all
>source-destination pairs across all 4 paths, but without regard to
>available bandwidth.  You can see that as follows where the load
>distribution used by CEF is ...
>
>Load distribution: 0 1 2 3 0 1 2 3 0 1 2 3 0 1 2 3 ,
>
>all lines being treated equally.  Is it possible to weight this
>distribution in a way that favors the higher bandwidth lines?  Or is
>there another alternative?

There is simple trick how to get unequal load-ballancing with CEF. Consider
you have two interfaces Serial0 and Serial1, and Serial1 bandwith is half of
Serial0. In static only scenario 2+1 ballancing can be achieved as

ip route 10.10.10.1 255.255.255.255 Serial0
ip route 10.10.10.2 255.255.255.255 Serial0
ip route 10.10.10.3 255.255.255.255 Serial1

ip route 172.16.0.0 255.255.0.0 10.10.10.1
ip route 172.16.0.0 255.255.0.0 10.10.10.2
ip route 172.16.0.0 255.255.0.0 10.10.10.3

Serial0 and Serial1 should also be configured to perform per-packet load 
ballancing:

int Se0
 ip load-sharing per-packet
int Se1
 ip load-sharing per-packet

In the case of BGP same technic could be used by replacing last three static
routes by corresponding number of BGP sessions.

>Thanks in advance for any suggestion.
>Earl
>
>----
>
>gw#show ip cef 1.2.3.4 internal
>0.0.0.0/0, version 90217, attached, per-destination sharing
>51032209 packets, 40384605152 bytes
>  via Multilink1, 0 dependencies
>    traffic share 1
>    valid adjacency
>  via Multilink2, 0 dependencies
>    traffic share 1
>    valid adjacency
>  via Multilink3, 0 dependencies
>    traffic share 1
>    valid adjacency
>  via Multilink4, 0 dependencies
>    traffic share 1
>    valid adjacency
>
>  51032209 packets, 40384605152 bytes switched through the prefix
>  tmstats: external 0 packets, 0 bytes
>           internal 51032209 packets, 40384605152 bytes
>  30 second output rate 998 Kbits/sec
>  High water mark 0, low water mark 0, current trigger set on no NHRP
>interface
>  Load distribution: 0 1 2 3 0 1 2 3 0 1 2 3 0 1 2 3 (refcount 1)
>_______________________________________________
>cisco-nsp mailing list  cisco-nsp at puck.nether.net
>http://puck.nether.net/mailman/listinfo/cisco-nsp
>archive at http://puck.nether.net/pipermail/cisco-nsp/


-- 
Regards,
Volodymyr.