Long Term Network Scenarios based on OBS/OPS

1. Long Term Network Scenarios based on OBS/OPS • 13 Partners: - Telecom Italia - Alcatel SEL AG - Alcatel CIT - Lucent Technologies Nederland BV - Marconi Communications ONDATA GmbH - Siemens • - Telefonica - FhG-HHI - IBBT - UCL - IKR - University of Stuttgart - UPC - ICCS Workpackage 3 Advanced Burst/Packet Switching [draft v1] Gert Eilenberger

2. Agenda The presentation is structured in two parts: • WP3 Objectives / Overview / Results • WP3 / 6 / 7 joint activities on optical burst switching nodes

3. Targeted Network Architectures PhoneHome PhoneHome Music Gaming Music Gaming Applications Applications SmartBiz SmartBiz Management Management Voice @Home Voice @Home Video Video Service Service TV User User Element Element Interactive Voice Multimedia Services Services Softswitch Control Control Network Capabilities User Location User profile Storage Resource Broker Multicast VPN L3 Packet L3 Packet Security Packet Network Services QOS Charging AAA DSL/FTTU Broadcast L2 Packet / Optical L2 Packet / Optical GigE / SAN SAN/NAS Optical Network Services G-MPLS Ethernet GRID LL Core Core Metro Access Access NOBEL WP3

4. Motivation for Burst/Packet Switching • Goal: Converged multi-service network with end-to-end QoS and multiplexing gain on network level Converged burst/frame switching network (new Layer 2 transport service) Premium Best effort Quasi 2 networks (packet switched) same technology stat. mux. Overprovisioning to get Best effort Premium unused premium quality premium Isolated best effort network Isolated premium network (packet) Status quo: 2 networks 2 technologies (circuit/packet) Premium Best effort Isolated premium network Isolated best effort network (pure TDM)

5. Motivation for Burst/Packet Switching (2) • Architecture options MSN Pure IP IP/OXC IP/DXC Data services TDM services Edge Routers Edge Routers Edge Routers Edge Routers IP router IP router Burstification unit Burst/Frame Switch (service agnostic) Optical Cross Connect Core Router SDH/SONET Cross Connect  Scalability  QoS  Costs  Integrated CP  Scalability  QoS  Costs  Flexibility  QoS  Opt. technology  Flexibility  Multi-layer control

6. Extended long-term scenario The new L2 network service with its hybrid circuit/burst/packet switching capabilities will be fully integrated into the GMPLS control plane (full vertical integration)

7. WP3 Objectives • Network and node architectures for high throughput optical burst/packet core and metro networks • Evolution from wavelength (circuit) switched to burst/packet switched optical networks: exploit improved statistical multiplexing • Exploit transparent opt. wavelength/burst/packet switching to reduce excessive electronic processing for reduced overall cost • Optimal balance between optical and electronic technologies in terms of performance and cost • NovelCP and MP functions adapted for optical burst/packet networks (performance monitoring, protection and restoration). • End-to-End QoS support in opt. burst/packet layer (reservation, allocation, signalling, signal regeneration etc.) • Possible extensions and/or evolution of standards

8. WP3 Key Achievements • Data Plane: Definition of requirements and traffic profiles for burst/packet networks and nodes • Data Plane: Various solutions for burst, packet and hybrid network architectures (dimensioning, performance) • Control Plane: Concepts on architectures and functions specific for burst/packet networks (routing, QoS, GMPLS) • Requirements and assessment of technologies for optical and opto-electronic burst/packet switching solutions • Evolution trends • … contained in >770 pages of deliverables

9. WP3: Advanced Packet/Burst Switching Deliverables • D4: “Requirements for burst/packet networks in core and metro supporting high quality broadband services over IP” (M6)  • D16: “Preliminary definition of burst/packet network and node architectures and solutions” (M14)  • D23: “Definition of hybrid opto-electronic burst/packet switching node structures and related management functions” (M20)  • D32: “Preliminary report on feasibility studies on opto-electronic burst/packet switching nodes” (M24) 

10. General Requirements and Characteristics

11. WP3 Migration Scenarios BS over dyn. l WR-OBS APSON ORION G.709 FS OBS/OPS Static l Dyn. l time Field deployment 2015 Product status Research lab status 2010 2005 “From semi-static to dynamically reconfigurable optical networks” technology

12. APSON: Adaptive Path Switched ON • Migration concept: via APSON to OBS/OPS networks New mechanism to support QoS: Just-the-Arrival-Time (JAT) reservation scheme

13. G.709 Frame Switching Concept • Aggregation of client packets into equally sized containers: G.709 Frames • Frame Aggregation Unit at network ingress and egress. • Switching of each individual G.709 frame. • Connection less forwarding. • Connection oriented bandwidth reservation and labeling. • Continuous G.709 OTUx connections on transmission links.

14. 3 1000 DSL users Installed access rates on 1GBit/s 1 concentrator link 2.5 DSL 0.8 Eth 10 Overdimensioning Eth 100 large web server, 2 LAN access 0.6 data center to WAN, Bandwidth efficiency 1000 users 0.4 1.5 super computers 0.2 talking to each other 1 100MBit/s 1GBit/s 10GBit/s 100GBit/s 0 0 2 4 10 10 10 Mean bandwidth on core link Aggregation factor Core Network Dimensioning Statistical Multiplexing in Core Networks • Aim: Reliable bandwidth estimation in packet based networks • Dimensioning rules for core links • Guidelines for the activity • Use knowledge of installed base andmeasurement of core link occupation • Typical provider knowledge • Avoid assumptions on user behavior and application mix. • Not predictable, outdated before consolidation • Expected result • Modified network dimensioning rules • independent of user behaviour • exploiting statistical multiplexing

15. Virtual Topology Design for OBS/OPS • Motivations for virtual topologies in OBS/OPS • Introduction scenario for OBS/OPS into wavelength-switched networks • Cost-optimal network design: reduce number of burst-switched interfaces by optical bypassing • Exploit lightpaths services for resilience and capacity adaptation • Combination of burst-switched and wavelength-switched networksin client-server hybrid optical network • But: dense virtual topologies also reduce statistical multiplexing gain  Integrative network design needed including effective contention resolution

16. Optical Burst Transport Networks (OBTN) OBTN Components • Use optical bypassing where possible • Allow constraint alternate routing • Assign shared overflow capacityfor alternate routes to improve statistical multiplexing(capacity share is defined by b) • Apply effective contention resolutionin nodes to achieve high QoS Summary • Reduction in burst-switched interfaces compared to OBS • Only small penalty in network resource efficiency • Overall high QoS Burst-switched trunk ports Comparison for COST CN network at 10-5 burst loss

17. ORION: Combining Packets and Circuits • ORION functionality • ORION node architecture

18. ORION: Emulation Results

19. Dimensioning of the two way reservation scheme for OBS networks (UCL) • Two way reservation scheme was introduced to avoid buffering in the core • OBS allows wavelength reuse (Reuse Factor RUF) • Network performance determined by RUF and possible utilization • Round-trip time tRTT and edge delay (for burstification) limit utilization • Example: with 10 ms edge delay, the network diameter should not exceed 1500 km

20. QoS provisioning in OBS networks • QoS provisioning in OBS networks: • Burst Length Differentiation for different traffic classes • Short Bursts (10 KB) for high priority data, real-time voice • Long Bursts (40 KB) for regular data • Extra Long Bursts (100s of KB) for fast data file transfer • Methods for QoS provisioning • Offset-Time Differentiation • Preemption window mechanism  most efficient for throughput and loss

21. Routing in OBS/OPS: Isolated adaptive connection-oriented routing • Adaptive path selection based on local node state • Congestion conditions • Actual link/buffer occupancy • Per packet decision, 3 algorithms studies • Path Excluding (PE), Multiple Choice (MC), Bypass Path (BP) • Improvement over simple Shortest Path (SP) NSFNet EON (COST266)

22. QoS routing in OBS networks • QoS concept based on Hamiltonian Path • Embedded ring topology to route Best Effort bursts • High-Priority bursts may use any (shortest) path

23. Clustering Architecture for Nodes of Optical Networks (CANON) • Clustering to reduce routing domain size for two-way reservation scheme  no buffering, low losses • Master nodes (MN) interconnected by mesh or rings over provisioned WDM channels

24. Control Plane Aspects • Adapting GMPLS to OBS/OPS networks • Current GMPLS protocol chain does not support CL OBS/OPS • WR-OBS (= fast ASON) maybe possible (Round Trip Time!) • Preventive resource reservation as work-around concept

25. GMPLS-UNI • GMPLS-UNI designed for CO Packet Switching • OBS/OPS with one way reservation need further extensions • Extensions developed for multi-layer interoperability of exisiting OIF UNI 1.0 • Single end-to-end signalling session (client-OTN-client) • GMPLS compliance for enanced QoS capabilities • Fast notification of failures (cross layer) • Explicit routing allows path diversity for protection • Failure recovery coordination (cross layer) • Scalability by information aggregation and cross layer information (reduced CP traffic volume)

26. Agenda • Taskforce “TCP over OBS”

27. TCP – Introduction • TCP (Transmission Control Protocol) is the dominating transport protocol in the Internet. • More than 80 % of the IP traffic today uses TCP on the transport layer. • TCP establishes an end to end connection: • Connection oriented • Reliability • Flow control • Congestion Control Application Oriented Layers L 5-7 (e.g. FTP) L 4 (TCP) L 3 (IP) L 2 (OBS) L 1 Transport Layer Network Layer Link Layer Physical Layer

28. Interaction of TCP and OBS TCP Delay, Delay Jitter Multiple Losses Reordering Aggregation of packets Losses (No buffering) “Aggregated” Losses Deflection Routing Buffering in the Node OBS

29. TCP Taskforce - Research Topics • Topics: • Impact of different TCP flavors and TCP parameters on TCP performance • Many TCP flows (highly aggregated traffic in metro and core networks) • Dependence of TCP performance on number of TCP segments of one flow in a burst • Performance of TCP with Deflection Routing • Applications: • FTP traffic (long-living TCP connection, bulk transfer) • HTTP traffic (short-lived TCP connections, short transfers ) • Scenarios: • Single Client/Server behaviour • Behaviour of many Clients and Servers

30. Impact of aggregation level on TCP performance blue: lossless red: burst loss rate 1% Application:Heavy Web Browsing x-axis: simulation time y-axis: throughput (bits / sec.) Summary: Higher aggregation level, i.e. higher number of aggregated clients, reduces the negative effect of burst losses on TCP performance, since less TCP segments per flow are affected by loss of a burst 3 Web Browsing Clients 300 Web Browsing Clients

31. Realistic Traffic model for TCP over OBS NOBEL Classical • The classical model considers only one TCP client and one TCP server. • The new NOBEL model considers one TCP client, one TCP server and additional traffic sources (fractal traffic). • The real number of TCP segments per burst from a single flow is lower than previously assumed. • Simulation with TCP SACK and Reno • Classical model (without additional traffic): An optimal value of the timer can be found. • NOBEL model (with additional traffic): Throughput is similar for the different values. • TCP SACK achieves higher throughput than TCP Reno.

32. Path B Path A Server Client TCP Performance with Deflection Routing • TCP is sensitive to Deflection Routing. • Deflection Routing is useful for contention resolution, as the performance degradation due to deflection routing is considerably smaller than the degradation due to burst losses. • The aggregation of more packets out of one TCP flow in a burst has positive impact on TCP performance with deflection routing.

33. Summary – TCP over OBS • TCP performs well over OBS networks, if an appropriate TCP parameter set is used and multiple TCP flows are aggregated into bursts. • MSS/MTU size heavily impacts the performance of TCP in OBS networks: high MSS/MTU values result in much lower effect on burst loss. • The advertized receiver window should be set to the maximum value in OBS networks • The higher the number of active users, the lower the effect of burst loss on the throughput in the network • TCP SACK achieves higher throughput than TCP Reno. • Real number of TCP segments from a single flow is lower than previously assumed • Deflection Routing has a negative impact on the TCP performance, but it is useful for contention resolution, as the performance degradation due to deflection routing is considerably smaller than the degradation due to burst losses.

34. Outlook for NOBEL 2 • New WP3 will collect work from old WP3, WP6, WP7 • Activity 3.1 „Architectures for future advanced burst/packet networks“ • Network and node architectures based on the opto-electronic solutions drafted by NOBEL WP3 • Optical burst/packet switching techniques for core and metro networks (reduced O/E/O) • Activity 3.2 „Control and management aspects of burst/packet networks“ • Control plane extensions specific to burst/packet techniques (inclusion in integrated GMPLS control plane) • Novel network control & management functions adapted for optical burst/packet networks • QoS support in the burst/packet layer (new layer 2 network services and service classes) • TCP-over-OBS Taskforce will be an important part of work • New TCP flavours • Traffic source models derived from measurements from NOBEL partners • Influence of new application mixes, traffic asymmetry and burstiness • Burst loss due to collisions (network load, topology, burst distribution) • Evaluation of hybrid solutions (Circuit / OBS) • Generalization of burst reordering problem in high-speed core networks

35. Optical Burst/Packet Switching Networks and Nodes Architectures Functional analysis Physical layer modelling and performance Technology/component aspects Workpackages 3, 6, 7 Joint Activities Gert Eilenberger

36. Joint WP3/6/7 Activities - Approach • WP3: • OBS/OPS network and node architectures • Functional analysis, control aspects, feasibility studies • WP6: • OBS nodes detailled architecture and implementation aspects • Physical layer modelling, performance analysis • WP7: • Technology and components options for OBS nodes • Technology/components depending performance limitations, e.g. crosstalk • Close links between WPs (common architectures…)

37. Joint WP3/6/7 Activities - Results • WP3/D16 • Various OBS node architectures (AWG, BAS, TAS), OPS architecture • First performance studies (X-talk, noise, no of , throughput) • Technology and component aspects (requirements, market view) • WP3/D23 • CANON architecture, improved AWG based node + performance • WP3/D32 • TAS nodes feasibility/performance (cascadability, eff. throughput) • AWG nodes feasibility/performance (BER vs. X-talk)

38. Joint WP3/6/7 Activities - Results • WP6/D24: • OBS node architectures based on cyclic AWGs (no results) [LUNL] • OBS nodes (Class I – III) with physical layer modelling and performance results (OSNR, filters, WCs) [ICCS] • WP7/D34: • Scalability and cascadability of OBS nodes (technology/component aspects): noise + crosstalk => max. no. of wavelengths • AWG based nodes, BAS nodes, TAS nodes

39. OBS node edge router Physical limitations: • Component availability • Signal degradation (BER) • Noise • Crosstalk • Nonlinearities Why physical OBS node design? Which throughput can be achieved with state of the art components?

40. Tune and Select (TAS) Node Architecture

41. 12 256 10 8 148 132 Throughput [Tbps] 6 96 4 2 10 0 0 NRZ NRZ NRZ RZ RZ RZ modulation format SOA type Gain-clamped SOAs have to be used! NRZ modulation is superior Throughput of TAS Nodes with 4 Fibersat 10 Gbit/s maximum throughput due to physical limitations maximum number of wavelengths per fiber allowed throughput to achieve a burst loss rate < 10-6 (effective throughput) Reference SOA Conventional SOA Gain-clamped SOA

42. BAS Opt. Packet Crossconnect Architectures Class-I Class-III • Applying tunable  converters

43. AWG Based OBS Node

44. Comparison AWG vs. BAS Architectures

45. WP3 Thank you for your attention!

46. Backup slides (will not be shown)

47. Connection-oriented OPS scenario • Main Property: Shared WDM links • Several wavelengths to choose from on the same output fibre • Problem: • Algorithm to map the Optical Virtual Circuits (OVCs) into the output wavelengths • Solution: • At OVC set up: Assign the OVC to the optimum wavelength • Using a dynamic wavelength assignment during the OVC life: In case of congestion, move the OVC to another wavelength using a Wavelength Selection(WS) algorithm

48. 10-1 TSWS LBWS SKWS 10-2 10-3 Packet Loss Rate (PLR) 10-4 10-5 10-6 0.4 0 0.5 1 1.5 2 2.5 Granularity D 1.2 QoS provisioning: Different WS Algorithm per Service Category • ATM like scheme: Provide K different categories of service based on K different WS algorithms • Each WS algorithm presents different performance • Thus, we can map the service categories into the WS algorithm according to the QoS requirements of these service categories • Case study: • 3 Categories of service • 3 WS algorithms • Problem: • The WS algorithms do not have performance alignment with the optical buffer granularity (D) • D = (FDL-size / Average IPpacket-size) x (Vt / Vp) • Our solution: • Redesigning the Optical Buffer architecture TSWS: Two State WS LBWS: Loss Bounded WS SKWS: Sequence Keeping WS

49. 1 RT LS 10-1 BE 10-2 10-3 Packet Loss Rate (PLR) 10-4 10-5 10-6 10-7 10-8 0 0.2 0.4 0.6 0.8 1 Granularity D QoS provisioning: Proposed Optical Buffer Architecture • Non consecutive FDL • FDL sequence: multiples of 1.2 / 0.4 = 3 • Example: • Optical buffer with 6 FDLs: Sequence: 0, 1, 2, 3 (1 x 3), 6 (2 x 3), 9 (3 x 3) • With this Optical Buffer Architecture we got: • The alignment of the WS algorithms performance • The aimed QoS provisioning: • Real Time (RT) • Very low PLR and no out of sequence packet • Loss Sensitive (LS) • Bounded PLR • Best Effort (BE) • Acceptable PLR

50. QoS provisioning: Proper Optical Buffer Architecture • Consistency of the solution: • Such a non-consecutive Optical Buffer Architecture depends on two design parameters, namely the propagation rate (Vp) and the transmission rate (Vt), an on the average IP (MPLS) packet size • The average IP packet size measured at the Catalan Academic Network over one day in September 2003 was 582 Bytes. • This measure done one year later (in October 2004) raised to 641 Bytes • And this year (in October 2005) we obtained an average IP packet size of 662 Bytes