1. Outline n IP over WDM u Motivations u Protocol stacks u Network architectures n IP/WDM integrated routing u Problem statement F Two-layer routing problem u Possible solution strategies F Integrated routing at IP and WDM layers Interaction with the routing protocols used in IP networks F Greedy distributed solution F Network-wide centralized solution n Extensions n Summary

2. IP over WDM - Motivations n IP traffic volumes u Traffic volumes on the Internet double every six months u Aggregate bandwidth required by the Internet in the US by the year 2005 is expected to be in excess of 35 Terabytes/sec n New high-capacity networks u To meet this anticipated need, carriers in the US are in the process of deploying high-capacity networks (OC-48~2.5 Gbps, and soon OC-192 ~10Gbps) for the sole purpose of delivering Internet data u Some new carriers are building networks customized for IP traffic (most existing “transport” networks were built primarily for voice traffic) n IP-centric and IP multi-service networks: Voice over IP, Video over IP,...

3. IP over WDM - Motivations n WDM reduces costly mux/demux function, reuses existing optical fibers. u Alternative to new fiber installation u Consolidation of legacy systems u Maximizes capacity of leased fibers u Future-proofing of new fiber routes n WDM allows high flexibility in expanding bandwidth n Cost Reduction - integrating optics and eliminating mux stages n Operation Efficiency - elimination of redundant protocol layers n Transport Efficiency - elimination of transport protocol overhead n Emergent technology is evolving WDM from optical transport (point-to- point line systems) to true optical networking (add-drop multiplexers and cross-connects)

4. IP: Internet Protocol AAL5: ATM Adaptation Layer 5 ATM: Asynchronous Transfer Mode SONET: Synchronous Optical NETwork PPP: Point-to-Point Protocol HDLC: High-level Data Link Control WDM: Wavelength Division Multiplexing SDL: Simplified Data Link provides length-based delineation instead of flag-based delineation WDM SONET/SDH ATM AAL5 IP 1 WDM SONET/SDH HDLC PPP IP 2 WDM SONET/SDH SDL IP 3 [1] W. Simpson, “PPP over SONET/SDH,” IETF RFC 1619, May 1994. [2] J. Manchester, J. Anderson, B. Doshi and S. Dravida, “IP over SONET,” IEEE Communications Magazine, Vol. 36, No. 5, May 1998, pp. 136-142. IP over WDM - Protocol stacks

5. IP over WDM - Network architectures With and without SONET/SDH multiplexing All three protocol stacks can be used in conjunction with SONET/SDH multiplexing Even without SONET/SDH multiplexing (for example R3 to R6 communication), since IP routers have SONET/SDH interfaces, IP over WDM could involve a SONET/SDH layer SXC ADM R WDM NE R1R1 R2R2 R3R3 R6R6 WDM NE WDM NE WDM NE R4R4 WDM NE SONET/SDH ring ADM SXC R7R7 R5R5 SONET/SDH Cross-Connect SONET/SDH Add-Drop Multiplexer IP Router WDM Cross- Connect or Add-Drop Multiplexer

6. Multiplex several SONET OC3, OC12, OC48 interfaces on to one fiber using WDM OC3/OC12/OC48 SONET/SDH HDLC PPP IP OC3/OC12/OC48 SONET/SDH HDLC PPP IP * Could even multiplex some IP/AAL5/ATM streams with IP/PPP/HDLC streams R R R R WDM Multiplexer WDM Multiplexer WDM SONET/SDH HDLC PPP IP IP over WDM - Network architectures

7. IP/WDM integrated routing - Problem statement Develop algorithms for integrated management of routing data in IP over WDM networks With SONET cross-connects, it becomes a three-layer problem With SONET cross-connects and ATM switches, it becomes a four- layer problem Problem space IP over WDM without multiplexing capabilities in intermediate layers 2-layer problem IP over WDM with multiplexing capabilities in intermediate layers 3 or 4-layer problem Solution space CentralizedDistributed

8. Two-layer routing problem R1R1 R2R2 R3R3 R5R5 R6R6 R7R7 R4R4 Virtual TopologyPhysical Topology R1R1 R2R2 R3R3 R6R6 R7R7 OXC R5R5 R4R4  What are the benefits/costs (in terms of network performance and management complexity) of performing traffic/QoS management and survivability at the WDM optical layer instead of at the IP layer?  Is there a hybrid or cooperative approach that is more optimal given a set of realistic performance and complexity constraints?

9. What is particular about this (IP/WDM) 2-layer routing problem? Limit on the number of optical amplifiers a lightpath can traverse before requiring electronic regeneration  All wavelengths amplified equally at an optical amplifier Without wavelength changers at OXCs (Optical Cross-Connects), wavelength assignments to lightpaths need to ensure availability of selected wavelength on all fibers on the lighpath R1R1 R3R3 R6R6 R7R7 OADM OXC R5R5 R4R4 R2R2 Optical Amplifier

10. Solution strategies Integrated routing at the IP and WDM layers  Interaction between existing routing schemes at the IP layer and this new integrated solution “Greedy” distributed solution  Monitor lightpath utilization and change allocations of lightpaths between pairs or routers accordingly Centralized system-wide optimal solution

11. Generic integrated approach (not specific to IP) Solve four sub-problems:  1. Determine virtual topology to meet all-pairs (source-destination) traffic  2. Route lightpaths on the physical topology  3. Assign wavelengths  4. Route packet traffic on the virtual topology Sub-problems 1 and 4 are equivalent to a data network design/optimal routing problem  Capacity assignments between routers are determined for a given traffic matrix  Flows are determined along with capacity assignments Metrics optimized:  Minimize costs  Subject to an average packet delay constraint  use M/M/1 queues and independence assumption to determine delay [3] B. Mukherjee, D. Banerjee, S. Ramamurthy, A. Mukherjee, “Some Principles for Designing a Wide- Area WDM Optical Network,” IEEE Journal on Selected Areas in Communications, Vol. 4, No. 5, Oct. 1996, pp. 684-696.

12. Routing protocols used in IP networks Link state based routing protocols, e.g., Open Shortest Path First (OSPF)  Currently OSPF Link State Advertisements (LSAs) mainly include operator-assigned link weights  Shortest-path algorithms used to determine routing table entries based on these link weights (Dijkstra’s, Bellman-Ford)  Example: Shortest path from R3 to R7 is via R4 and R5 R1R1 R2R2 R3R3 R5R5 R6R6 R7R7 R4R4 1 2 1 4 1 11 3

13. QoS extensions to OSPF Flow-based IP traffic  Have LSAs include “available bandwidth”  Each flow has a required bandwidth; delete all links in graph that do not have requisite available bandwidth  Then apply shortest-path algorithm using link weights Connectionless traffic  Modified Bellman-Ford to determine shortest-paths using link weights  If there are multiple paths with the same minimal weight, then the path with the maximum available bandwidth is chosen [4] R. Guerin, S. Kamat, A. Orda, T. Przygienda, D. Williams, “QoS Routing Mechanisms and OSPF Extensions,” IETF Internet Draft, 30 Jan. 1998, draft-guerin-qos-routing-ospf- 03.txt.

14. Classification of routing schemes Optimal schemes base routing decisions on all-pairs source- destination traffic e.g., the integrated four sub-problem solution Shortest-path schemes make routing decisions for per-nodepair traffic e.g., OSPF [5] C. Baransel, W. Dobosiz, P. Gewicburzynski, “Routing in Multihop Packet Switching Networks: Gb/s Challenge”, IEEE Network Magazine, 1995, pp. 38-61. Routing schemes Table-basedSelf-routing Shortest-path routing (user-level optimization) Optimal routing (system-level optimization)

15. Interaction between OSPF and integrated solution No conflict:  The integrated solution changes “maximum” capacities between routers  OSPF (with QoS extensions) uses this information along with “available” capacities to make routing decisions Potential conflict:  Should the integrated solution change the forwarding table entries based on flows computed as part of the capacity assignment problem?  If so, both OSPF and integrated solution are changing forwarding table entries Other issues:  OSPF LSAs need to exchange maximum bandwidths  Can instabilities result in forwarding data if both OSPF and integrated IP/WDM routing software make changes?  What is the time scale of operation for the integrated IP/WDM software?

16. Greedy distributed solution WDM network routing does not change the virtual topology It measures utilization on each lightpath (between pairs of routers)  If under-utilized, decrease number of lightpaths or data rates used on lightpaths  If over-utilized, increase number of lightpaths or data rates used on lightpaths Using wavelength availability and optical amplifier related constraints, find shortest path for lightpath and establish crossconnections (“greedy” user-level optimal) Basis: optical layer routing should not change IP-layer routing data R1R1 R2R2 R3R3 R5R5 R6R6 R7R7 R4R4 R1R1 R2R2 R3R3 R6R6 R7R7 OXC R5R5 Virtual TopologyPhysical Topology R4R4

17. Centralized network-wide solution In greedy distributed solution, there may be instances when a lightpath could have been accommodated if routes or wavelength assignments of existing lightpaths had been adjusted All-pairs traffic demand is given; find optimal routes and wavelength assignments of lightpaths (also called the RWA problem) Network Management System R1R1 R2R2 R3R3 R6R6 R7R7 OXC R5R5 R4R4

18. Extensions Consider multiple QoS metrics while finding optimal solutions  For example, in integrated solution, consider packet loss ratio, packet delay variation, improved packet delay formulations (assuming MMPP traffic) Extend solutions to allow for multiple service classes  Differentiated services in IP networks  Simple schemes for packet tagging, classification and per-hop behavior  Integration of IP service classification with routing and wavelength assignment Allow for network and service survivability  Use full capacity or have spare capacity  Use protection fibers for increased throughput, but when fault occurs, throttle back best-effort traffic and accommodate all higher- priority traffic

19. Summary Defined IP over WDM network architectures and protocol stacks Defined routing problem statement for two-layer networks  Special features of WDM networks: optical amplifier constraints, wavelength continuity constraints Proposed three solution strategies:  Integrated IP/WDM optimal routing to operate in parallel with OSPF shortest-path routing  Greedy distributed solution - monitors traffic offered to WDM network and determines shortest-paths meeting certain constraints (user-level optimal)  Centralized system-wide optimal solution - adjusts existing lightpaths if needed to accommodate newly requested lightpath Identified possible extensions