MPLS 2009 – Day 1 Report

Fasten your Seatbelt!!!

Hello fellow techies.  If you are hungering for a rich meal of multicast VPNs with some MPLS-TP on the side, and MPLS in the Access for dessert – you have come to the right place.  These were the tutorial sessions today as the MPLS 2009 conference in Washington DC, and they were quite filling. In fact, I’m stuffed.

Let’s start with MPLS-TP.  Matthew Bocci and Lubo Tancevski from Alcatel-Lucent shared the near-impossible task of covering this topic in 1.5 hours.  They went through their slides extremely fast, such that even experts in this field would have missed quite a bit of what was said.  What I did catch was quite informative.  Much of the discussion was around defining what MPLS-TP is, or is not.  There were a few ideas floating around:  the presenters said that MPLS-TP is a subset of MPLS that is geared toward transport and not services.  Kireeti Kompella from Juniper brought up the point afterward that MPLS started as a transport mechanism, then we spent the last 10 years building services for it, and now we are stripping it back down to do just transport.  I must say he makes an excellent point, but everything in this industry comes full circle.  This is no different.

The presenters also spent a bit of time on OAM functions for MPLS-TP.  There are 3 ways to do OAM for MPLS-TP:

            Hop-by-hop through the control plane

            Out-of-band where replies come through a different path

            In-band – creating an associated channel

One of the requirements for MPLS-TP is that it be able to operate without a control plane.  Basically that means static provisioning through an NMS – the way typical SONet and OTN networks are provisioned today.   If it does operate with a control plane, it will be GMPLS.

Other areas of discussion include resiliency and protection using Protection Switching Coordination (PSC) protocol; use cases including Business VPN services, Carrier’s carrier, and mobile backhaul; and what’s going on with the standards.

All in all, an excellent presentation, I just wish we had 3 hours to cover it.

Up next was the equally speedy Multicast VPN technologies and Real World Deployments from Azhar Sayeed from cisco, and Maria Napierala from at&t.  They offered some great information on how multicast should work in MPLS to support MVPN, but much of the discussion was about using PIM vs. BGP in the MPLS network.  These guys were definitely the pro-PIM camp and every other slide was justification why PIM is better than BGP.  Of course I knew this would knot Juniper’s panties in a bunch, and sure enough the first person to the microphone after the presentation was Kireeti, and the second was Ina Minei.  Both said they felt the presentation was biased, and Maria challenged them to tell her where her findings were in error. 

The discussion also included the signaling component for P2MP and MP2MP LSP’s.  Basically RSVP-TE can only set up P2MP and if you want to do MP2MP, you must set up a full mesh of P2MP.  All the traffic engineering benefits still apply, so use this if you want hard QoS guarantees and FRR.  MLDP is a much simpler solution for setting up P2MP and MP2MP. 

Maria really hit the nail in the coffin with slides titled:  If PIM already exists and works, why use BGP.  Much of her presentation was regarding real-world architectures from at&t’s network and customers which shows deployments using GRE tunnels.  Her slides had a lot of great animations – too bad they don’t show up in the pdf version distributed by the conference. 

Another job well done, and again I wish the presenters had 3 hours to cover this topic. 

Last but not least was the MPLS in the Access topic by Kireeti from Juniper.  Apparently he had discussed this topic at last year’s conference, but felt that enough had changed and that it was important enough to revisit the topic this year. 

I enjoy Kireeti’s style of presentation and he is always a wealth of information.  His presentation introduced a whole new slew of terms (like we all need more acronyms to understand) and his main point is that the network should not determine which services can be offered, but the network should support any services that are offered.  So if you decouple the network from the services, you can place the service nodes (SN) anywhere you want.  You can centralize them, or you can push them out closer to the customer.  The Access Node (AN) is the first or last node in the provider’s network to handle packets from the customer at layer 2 and up.  The Transport Node (TN) just transports packets between AN and SN and SN and SN.  One other type of node was defined – the Service Helper Node (SH).  This node helps provide the services without ever touching the customer packets.  Examples of this might be a SIP server or Authentication device.

Other points in his presentation went through the challenges of having different network regions using different technologies, hence the argument for having MPLS end-to-end.  Some of the features required to do this are multi-segment pseudo-wires (MS-PW) and LSP stitching using labeled BGP.

Too bad Kireeti’s topic was at the end of a very full day. He too could have used 3 hours, but by this time my brain was full.

Generally speaking it was a great day, lots of rich nourishment for the MPLS-mind.  Stay tuned for my report on the sessions tomorrow which include:

MPLS Scaling

Fast Service Restoration for VPLS

SAN services over MPLS

Future Vision on Seamless Interconnection