[nsp] ip load-sharing per-packet - cef accelerated ?

[Dave Spencer]

On Mon, Mar 10, 2003 at 03:58:53PM -0800, Bruce Robertson wrote:
> In my opinion, MLPPP is a much better solution to the problem.  Yes there's
> a bit of a CPU hit, but with two T1s, for example, you now have a true 3 Mb/s
> pipe, not two 1.5 Mb/s pipes that are load balanced.  If a customer *needs*
> say, 2 Mb/s, you can provide it.  We use MLPPP everywhere in our network,
> and it always produces better results than load balancing.

Unfortunately, this doesn't work out for services delivered on a variety of
encapsulations.  We do MLPPP for HDLC customers, but for Frame customers we're 
limited to per-packet load sharing.  Our NxT1 ATM customers generally use IMA, 
which has its own issues to deal with, but works out cleanly as a single PVC
on our agg router.  Basically, we'll use whatever is necessary to solve any
given problem.

(Then there are the problems with legacy routers which can't _handle_ the CPU
hit...)

-Dave