Fast Reroute without MPLS?

Fast Reroute without MPLS?

I’ve been a long-time advocate of MPLS, because it enables simplification: You can eliminate your ATM and Frame Relay infrastructures and consolidate your virtual circuits onto your routed IP infrastructure.
I also like the separation of control plane and forwarding plane in MPLS, allowing you to simplify the core by distributing the complex “intelligence” components around the edges, leaving the core to do simple high-speed switching.
The watchword in all this, as it always is for me, is simplicity, which almost always translates into lower operational expense and lower operational risk.
One of the supposed benefits of MPLS is MPLS Traffic Engineering. You can use MPLS-TE to specify constraints that distribute your traffic flows across your network in much more elaborate ways than you can with simple IGP shortest paths.
But MPLS-TE is complex. It adds a signaling protocol (RSVP-TE) to your network that must be managed, and optimizing the various constraints is complicated. And mistakes with MPLS-TE can be dangerous.
For these reasons the only MPLS-TE capability that is widely deployed is Fast Reroute (FRR), which does provide a significant enough benefit for operators to run RSVP-TE signaling and the supporting IGP extensions in their network.
When a node or link fails in a traditionally routed network, the following must take place for the network to recover (assuming the network is running a link state routing protocol):
(1) The routers adjacent to the failure must detect the failure.
(2) The adjacent routers must inform the rest of the network about the failure.
(3) The routers in the network must calculate a new path bypassing the failure.
(4) The network must fully converge on the new path (that is, all routers must finish their recalculations and update their forwarding tables, so that no loops or black holes remain).
The problem is that these four steps can take several seconds or longer, which is a very long time to be suffering packet loss in a high-performance network core. Particularly if you have SLAs backed up with financial guarantees. SONET rings can significantly reduce this recovery time by switching around the failure, but they’re expensive.
This is where FRR comes into play: In an MPLS-TE network, FRR pre-calculates shortest paths around individual nodes and links so that if a failure occurs, traffic can be quickly switched to the reroute path: often as quickly as 50 ms. FRR does not take into consideration any best-path parameters, and its reroute paths are intended only as short-term detours around a failure while the network’s IGP calculates a new route.
However, it is feasible to implement a Fast Reroute mechanism natively in an IP infrastructure without using MPLS-TE.
Mike Shand and Stewart Bryant of Cisco Systems have had an Internet-draft in front of the IETF for a number of years now called IP Fast Reroute Framework that proposes an FRR solution without MPLS-TE. For operators who use MPLS only for the FRR functionality, this solution could be promising for simplifying their networks.
Certainly there are some questions to be answered: Will the solution scale? Does it go against the trend of moving intelligence to the edge and put complexity back in the core? (And if so, does that matter to the network operator?) Will it negatively affect traffic flows, or the performance of the IGP?
I’m looking forward to seeing where this proposal leads.