Re: IPsec & Markers

On Oct 8, 2004, at 9:43 PM, Michael Krause wrote:

>
>  I hate to bring this up but from a process perspective, did we not 
> already
> achieve consensus that markers would always be required?  We are
> getting towards last call so it would be good to avoid re-opening a 
> topic
> if it has already reached consensus.
>
>  Mike
>
Agreed.

The optimization that has been suggested is that MPA Mode bits be
allocated to deal with the case where both ends have an MPA-Aware
IPSEC implementation and that the MPA layer is aware the IPSEC is
being used (reliably) and the IPSEC mode is such that all PMTU
changes will be hidden by IPSEC (i.e. the virtual path's PMTU
will never change).

Not only do I think the above scenario is unlikely, but if your
platform is doing IPSEC in the same device as RDDP/RDMAP
and MPA/TCP then why are you worried about the CPU overhead
of calculating markers?

In my estimation for every 100 connections where this condition
could apply the "savings" would be significant for at most one.
More importantly there will be billions of  connections where
these option bits are totally irrelevant.