A Unified Control Plane Dream or Pipedream? Adrian Farrel Old Dog Consulting IETF Routing Area Director Old Dog Consulting Agenda • History • Where does it all come from? • Objectives and Dreams • Development • Extensions and Divergence • Success Stories • Disappointments • Why are we here? • Why has GMPLS not taken over the world? • Why are we here in Tokyo, now? • Where are we going? © Copyright Old Dog Consulting 2010 Page 2 MPLS is Established • MPLS is 13 years old • I have the T-shirt from MPLS2007 • All (nearly all?) major service providers have MPLS in their core networks • The majority is LDP or L3VPN • MPLS-TE has limited, but successful deployment © Copyright Old Dog Consulting 2010 Page 3 WDM and Automation • As MPLS-TE was being developed • Technical advances in WDM • Deployment and research of WDM systems • Management-based solutions becoming complex • Proposals extend MPLS-TE to provide an automated control plane for WDM systems • Multiprotocol Lambda Switching (MPλS) • draft-awduche-mpls-te-optical-00.txt • April 2000 • Awduche, Rekhter, Drake, Coltun © Copyright Old Dog Consulting 2010 Page 4 Generalisation • If you can do it for packets and lambda • Why not do it for all connection-oriented networks? • Isn’t all circuit switching the same? • CO-PS • • • • MPLS ATM Frame Relay Ethernet • CO-CS • Fibre • Lambda • TDM • draft-ietf-mpls-generalized-signaling-00.txt • October 2000 • Multiple ideas, many authors © Copyright Old Dog Consulting 2010 Page 5 Protocol Development • MPLS-TE protocols had been developed already • Routing • Signalling • Generalisation of these protocols to GMPLS • Intent that GMPLS included all existing traffic engineered MPLS • New protocols only where new needs • LMP • PCEP © Copyright Old Dog Consulting 2010 Page 6 Two Protocols for Every Use • • • • OSPF and IS-IS RSVP-TE and CR-LDP LMP and Nortel’s own offering Lessons from history • • We do not need multiple solutions for the same problem Development cost is more than doubled • Who pays? • Interoperability is compromised • Providers are “locked in” • Deployment is complicated • Additional or more expensive operations teams • Company mergers, etc. become a nightmare • “Wrong” decisions are made • Consider regional standards that pick the “wrong” protocol Two roads diverged in a yellow wood, And sorry I could not travel both And be one traveller… Robert Frost (The Road Not Taken – 1915) • The IETF made decisions • • • Only LMP was taken into the CCAMP working group RFC 3468 stopped work on CR-LDP OSPF and IS-IS too well deployed to make a choice © Copyright Old Dog Consulting 2010 Page 7 And Did It Work…? • Lots and lots of implementations • This was around 2000 • Everyone was building an optical switch • Most implementations were for WDM • Significant research • Theoretical work to prove utility of control plane • Experimental equipment and networks • A lot of successful control plane interop testing • GMPLS-enable equipments shipped © Copyright Old Dog Consulting 2010 Page 8 Where are the Deployments? • GMPLS deployments do exist • WDM deployments tend to be small • Metro add/drop • GMPLS is a management tool • Reduces the complexity of provisioning • Single-touch connection set-up • Network status information gathering • Intelligence remains in the NMS (not in the network) • Some significant long-haul networks • Networks tend to be very stable • GMPLS is just a provisioning tool • SDH networks • Many networks deployed • Some are quite large • GMPLS has not taken over the world © Copyright Old Dog Consulting 2010 Page 9 And What Didn’t Work? • Many issues conspired to slow GMPLS • The bubble burst • TDM deployments too established • Retro-fitting GMPLS not attractive • No pressure to migration packet networks from MPLS-TE to GMPLS • PBB-TE didn’t take off • But GMPLS did fulfil its technical promise © Copyright Old Dog Consulting 2010 Page 10 Why Isn’t GMPLS Widely Deployed? • • • • The equipment was available The providers were looking at deployment But? There are a number of roadblocks • • • • • • Data plane interoperability Equipment cost Control channel interoperability Control plane interoperability Operational hurdles Network complexity © Copyright Old Dog Consulting 2010 Page 11 Roadblocks 1: Data Plane Interop • No point in a unified control plane if the data plane doesn’t interwork • WDM systems • • • • Different choice of lambdas Different power levels Complex optical impairments Different encodings • TDM • SONET/SDH • Different options and features • 2.5G, 10G, 40G encoding and modulation • Why? • Uncoordinated development under time pressure • Regional preferences • It is not a benefit to the incumbent vendors to interoperate © Copyright Old Dog Consulting 2010 Page 12 Roadblocks 2: Equipment Cost • GMPLS is most effective in dynamic networks • Dynamic networks need flexible, reconfigurable equipment • Flexible equipment has been expensive • For example, in a WDM network • Best flexibility is achieved using OEO • OEO has been the most expensive equipment • For example, control plane handling • Previous transport equipment • Only needed lightweight CPU • Only needed low bandwidth management channels • Introduction of a control plane added cost © Copyright Old Dog Consulting 2010 Page 13 Roadblocks 3: Control Channel Interop • Mainly an issue in WDM systems • How to adjacent WDM nodes communicate in the management and control planes? • In the lab we use 10/100 Ethernet • You can’t deploy that • You could connect to an IP cloud • Most WDM equipment has an Optical Supervisory Channel (OSC) • There are no standards for the OSC • It is not a benefit to the incumbent vendors to interoperate © Copyright Old Dog Consulting 2010 Page 14 Roadblocks 4: Control Plane Interop • The point of standards is to achieve interoperability • Multiple conflicting standards do not help • Why does “standards shopping” happen? • Vendors want to add value • Competitive edge is important • Vendor-specific extensions tend to break interoperability • They don’t need to • Why should an incumbent vendor enable interop? © Copyright Old Dog Consulting 2010 Page 15 Roadblocks 5: Operational Hurdles • Transport network operation is well established • Transport operators are conservative • Risk-averse • Demand stability • Huge investments already made • Extensive management systems • Education and training • A control plane is a big hurdle © Copyright Old Dog Consulting 2010 Page 16 Roadblocks 6: Network Complexity • GMPLS is intended to simplify the network • Why do people think it makes it more complex? • A very rich function set • Core GMPLS includes many features that are “advanced functions” in traditional networks • A very advanced toolkit • We are engineers – we like to build things • It is easy to apply GMPLS to some very complex problems • Vendors need to understand and sell simplicity • Service Providers have to learn to prioritise their requirements © Copyright Old Dog Consulting 2010 Page 17 So, Why Are We Here? © Copyright Old Dog Consulting 2010 Page 18 The Prospects Are Still Good! • Plenty of other reasons to be here • • • • • • • Interworking between network “islands” Continuation of the Ethernet project Optical Transport Networks (OTN) Advances in WDM Green Networking Integrated networking (IP-over-Optical) MPLS Transport Profile © Copyright Old Dog Consulting 2010 Page 19 Network Interworking • When there are vendor islands • Still want end-to-end automation • Need “service interface” (UNI) • Need glue between networks (E-NNI) • ITU-T ASON architecture makes these clear • MEF calls specifically for a UNI • Huge benefit in a standard protocol solution at these interfaces • ITU-T solutions (PNNI, CR-LDP, RSVP-TE) • OIF solutions (RSVP-TE) • GMPLS • We need a solution • • • • We don’t need five solutions! If GMPLS is also used at (some) I-NNI then choose GMPLS All the RFCs exist for immediate deployment for ASON I-Ds to support MEF UNI are about to become RFCs © Copyright Old Dog Consulting 2010 Page 20 Ethernet TE is Not Dead • The scope of PBB-TE is not as large as predicted • Campus-style deployments are still likely • Core backbone usage to link routers • Not used to build a fully-meshed core • Network diameter and complexity is not huge • Somewhat complicated resource sharing required • Planned reduction in forwarding table size • GMPLS offers automated and simplified management • I-D is about to become and RFC © Copyright Old Dog Consulting 2010 Page 21 OTN • G.709 is not new • Sub-lambda technology • Recent major advances in technology • New revision of G.709 (version 3) • Support for 1.25Gbps and 2.5Gbps • ODU-flex • More flexible and attractive • Considerably interest in implementation and deployment • RFC 4328 • Support for G.709 version 1 • New work in CCAMP • Re-assess label format • Support all resizing and advanced features © Copyright Old Dog Consulting 2010 Page 22 WDM Resurgence • Continued increase in bandwidth demand • Introduction of lower-priced components • PICs make OEO more affordable • Introduction of smaller all-optical crossconnects • 2x2 and 4x4 matrices • Makes phased deployment of PXCs realistic • ROADMs • GMPLS building blocks all in place • Next steps are impairment-aware routing • First stages almost complete in CCAMP and PCE © Copyright Old Dog Consulting 2010 Page 23 Green Networking • A very real demand to reduce energy use • Requirements are not limited to equipment • It is important to route traffic to • Use most power-efficient path • Make best use of existing paths • Increases the pressure for advanced TE • Needs to be dynamic • Needs sophisticated path computations • Most effective when integrated across layers © Copyright Old Dog Consulting 2010 Page 24 Integrated IP-Optical • This is the sixth year of iPOP • Are we wasting our time? • Maybe operators really don’t want this • Too dynamic • Too hard to operate • Too complex to deploy • Many attractions • Flexible equipment deployment • Flexible re-grooming • consolidated operations • Facilitated by many innovations • High capacity, tuneable, interfaces on routers (lambda, OTN…) • OTN flexibility • Plug-and-play integrated devices • Advanced planning software and PCE • Integrated GMPLS control plane © Copyright Old Dog Consulting 2010 Page 25 MPLS-TP • Transport-grade MPLS • • • • • • • • OAM Bidirectional Protection and restoration Optional, high-grade, TE control plane Work in the IETF with ITU-T cooperation Control plane will use GMPLS Equipment interoperability is a MUST Questions: • Will a control plane be used, or just management? • Will GMPLS be adopted in MPLS-TE networks? © Copyright Old Dog Consulting 2010 Page 26 Where Next? • A rocky road, but… • We have a very rich control plane toolset • The future is in your hands • We have all of the building blocks • New work is either very specific or very minor • We have the experimental evidence • The vendors have a marketing story • The providers see the benefits • Get on with it! • Issue the RFQs • Build and ship the products • Deploy the networks © Copyright Old Dog Consulting 2010 Page 27 Questions adrian@olddog.co.uk adrian.farrel@huawei.com © Copyright Old Dog Consulting 2010 Page 28