Seems to me the original poster was correct. The only effective method of
load balancing would to either except defaults via an igp from the upstream
or to just default them statically. If you wanted control over what traffic
exits which interface you could add statics to specific providers. Since
the default route is the least specific it would follow the more specific
route.
For safety sake, in the event of a interface failure the static should be
pointed to the interface, not the remote gateway. That way when/if the
interface fails the default will go away and all traffic will be routed to
the live interface. But, for sure, per packet load balancing would cause
problems.
-David
Network Engineer
-----Original Message-----
From: Tony Tauber [mailto:ttauber@genuity.net]
Sent: Friday, October 13, 2000 11:09 AM
To: Cliff Judge
Cc: Jose Ferreira; 'cisco-nsp@puck.nether.net'
Subject: Re: [nsp] Multi-Homing without BGP
On Fri, 13 Oct 2000, Cliff Judge wrote:
>
> I would suggest balancing on a per-packet basis in this kind of situation
> (i.e. don't enable ip route-cache on the interfaces)...you cannot control
> the routes you are placing into your cache this way, so it is best to
> throw the packets evenly out both interfaces.
Excuse me?!? Why is it best?
I'll tell you why it's worse:
Chosing between two providers to some distant location the delay
is liable to vary considerably thus, there's a high probability
of packets arriving out of order which will likely cause
retransmission requests from the destination IP stack which
thinks it's missing intermediate segments.
Excessive retransmissions will not only waste bandwidth but slow
effective data transfer between application layers significantly.
I'm from the camp of thinking that originating your routes
via BGP from your own AS is the cleanest way to multihome;
however, if you're just recieving some sort of default from
your upstreams, all above still holds.
Tony
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:18 EDT