MPLS 2009 – Day 2 Report
Before I comment on the technical sessions today, I just want to let everyone know that they did have power to about ½ the tables today, and they did have decaf coffee. I know you were worried, but alas the conference administrators came through.
While this is day 2 of the conference, it is really the start of the general sessions so the kickoff was this morning. The themes today were MPLS-TP, Seemless and Scalable MPLS, and High Availablity MPLS with a little bit of cloud computing and SAN services thrown into the mix.
The keynote speech was given by Raymond Zhang from BT on the topic of cloud computing. He asked the question: “what if 90% of customer applications were in the network?” How would we handle it? He also made a comment that made me chuckle a little. He said that in the networks we build the infrastructure to treat faults as exceptions, but software is written to deal with faults as the norm. HA! I’m guessing he must be a windows user.
Next up was Yakov Rekhter from Juniper to discuss Local protection for LSP tail end node failure. Yakov was informative as usual, and had excellent answers to the problem of “what happens when the end node (PE) fails?” We have all kinds of mechanisms to deal with failures in the middle of the network, but that last node is the bugger that will kill the service if it dies. One key point he made was that High Availability (HA) can NOT depend on routing convergence. It must be independent. The only way for HA MPLS networks is for fast recovery.
He re-introduced the concept of LSP hierarchy (which I heard about at this conference a couple years ago) and stated that connectivity recovery options must support this concept of hierarchy. So if the PE fails, the next hop P is the Point of Local Repair (PLR), and it must redirect traffic to a different PE. The recovery PE must know what labels the original PE assigned, so it must have a different LFIB that is a mirror image of the primary. He called this a “context-specific” LFIB.
Kireeti Kompella from Juniper was next and gave his talk on scaling MPLS to 100,000 nodes. Many of his slides were the same as yesterday discussing the connectivity blueprint and the new terminology of Access Nodes, Service Nodes, etc. His 2 talks were definitely related. If MPLS pushes further out into the access network, then it has to scale. His examples were MPLS to the DSLAM and MPLS to the cell site. Think about how many cell sites there are and it automatically hits you that we have a scalability concern.
Nicolai Leymann from Deutsche Telecom was the service provider sanity check on this whole concept of scalability and seemless MPLS. They are doing MPLS in the access and pseudowires from AN to the SN. He said that the key is to separate services from transport, and in doing so you can move the services anywhere they need to be in the network.
Next was George Swallow from cisco on MPLS-TP and blurring the boundaries. One comment he made was MPLS-TP not a radical new change, but a significant game changer. Another point he made was if pseudowire service is end-to-end, then maybe we want MPLS-TP end-to-end as well, and if so, where is the boundary between packet and transport in this picture? Very blurry.
After lunch was the Ethernet theme. Andy Malis from Verizon kicked things off by discussing Verizon’s requirements for carrier Ethernet. I had a chance to talk with him before his presentation and he made it clear to me that Verizon is doing PBB and MPLS-TP. PBB-TE (PBT) is not a strategy for Verizon. He went through the importance of the MEF standards as a common way to define services. He also addressed the scalability comparison between H-VPLS and PBB+VPLS – guess who wins?
Luca Martini from cisco discussed his 10 years of pseudowires – where they came from and where they are going. He commented on the fact that the standards bodies take so long to get things standardized, that his draft went through 153 revisions before it became RFC 4447. RFC’s for multi-segment pseudowires are just about to come out, yet vendors have been supporting these pre-standard features for some time now. The market does not wait for standards.
Other discussions today include Fast Restoration for VPLS, SAN services over MPLS, New Internet Architecture, Path Computation, and to wrap up the day Monique Morrow from cisco moderated a 10-person panel on emerging technologies.
Another very full day – hope I have some brain capacity left for tomorrow’s sessions.
