--- Begin Message ---
Hi all,
I wonder if anyone has recently compared *-in-IPv6 against SR MPLS, for new
service provider network deployments (where you have the typically requirements
of L2 MPLS VPNs, L3 MPLS VPNs, some sort of label distribution method in the
underlay, and a BGP free core).
There are a few problems with MPLS, and rather than trying to fix MPLS,
removing MPLS could be nicer, and I think IPv6/4/Ethernet-in-IPv6 maybe
achieves that.
The problems I have with MPLS are:
* Label space is limited to ~1M labels/20 bits -> IPv6 has ~128bits of space
* MPLS only has 3 bits for EXP -> IPv6 means no changes or "mapping" required
from DSCP to EXP and back to DSCP
* MPLS has no payload protocol ID field -> IPv6 has a next header field, which
removes the need for PW CW and the broken load-balancing we see today
* MPLS required an entropy label + ELI, or FAT label for better load-balancing
-> multiple IPv6 addresses can be assigned to the same FEC to improve load
balancing
Has anyone here seriously looked into *-in-IPv6 instead of rolling out SR MPLS,
if yes, what did you find?
(Note: I discount SRv6 because that is very different to *-in-IPv6 and, it just
introduces a whole new bunch of problems).
Cheers,
James.
--- End Message ---
