All Rights Reserved © Alcatel-Lucent 2008 (G)ELS - Ethernet VLAN-label Switching (ELS) (G)ELS - Ethernet VLAN-label Switching (ELS) Benchmarking Carrier Ethernet Technologies Workshop Session MII.1 Krakow, Poland April 30, 2008 Dimitri Papadimitriou

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Evolution of Ethernet paradigms Two main scalability concerns:  VLAN ID space - can not be solved with Provider bridges (IEEE 802.1ad)  MAC address space & learning - (hierarchical) hash-based table lookup (=> simple but limited MAC table size due to memory consumption & non-deterministic lookup time) Main “networking” concern:  Loop avoidance (STP) - can not be solved with STP 802.1d or RSTP 802.1w  Convergence time of STP - idem Main performance concern:  STP “blocks” network trunks - not solved with MSTP 802.1s Spanning Tree Protocol (STP) (VLAN-)Bridges Multiple STP (MSTP) Provider Bridges (PB) Multiple STP (MSTP) Provider Backbone Bridges (PBB) Ethernet LAN/MAN bridging branch

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Ethernet MAC and Ethernet v2 Ethernet MAC Frame Ethernet v2 Frame Preamble SD Destination Address Source Address Length Information Pad FCS to 1518 bytes Synch Start frame MAC address (6 bytes) is either Single address (0x0….) Group address (broadcast = ) MAC addresses are defined on local (0) or global (1) basis (second bit) 2 46 possible global addresses Preamble SD Destination Address Source Address Type Information Pad FCS to 1518 bytes Synch Start frame 6 6 No TTL (time to live) => impossible to detect looping Ethernet MAC frames

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Spanning Tree Protocols: Count-to-infinity Spanning tree: a connected, acyclic subgraph (no cycles) containing all the vertices of a graph Minimum spanning tree (aka shortest spanning tree): a weighted graph which contains all of the graph's vertices Count-to-infinity problem (as for any other Distance-Vector routing protocol)  Temporary forwarding loop (cycle) that con persist for O(10s)  (R)STP does not provide for fast convergence (and no - known - suitable technique to improve distance vector convergence properties) Note: steiner tree = a minimum-weight tree connecting a designated set of vertices, called terminals, in an undirected, weighted graph or points in a space. The tree may include non-terminals. Source: Dictionary of Algorithms and Data Structures [online], Paul E. Black, ed., U.S. National Institute of Standards and Technology. 17 July RootRoot unreachable cycle

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Evolution of Ethernet paradigms: PBB Two main scalability concerns:  VLAN ID space - solved: S-VID (12 bits) -> I-SID (24bits)  MAC address space & learning - solved: MAC-in- MAC tunneling (MAC learning still required) Main “networking” concern:  Loop avoidance (STP) - not solved  Convergence time of STP - not solved Main performance concern:  STP “blocks” network trunks - not solved Spanning Tree Protocol (STP) (VLAN-)Bridges Multiple STP (MSTP) Provider Bridges (PB) Multiple STP (MSTP) Provider Backbone Bridges (PBB) Ethernet LAN/MAN Bridging branch

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Ethernet Transport technologies SwitchingBridging Ethernet ELS: Ethernet VLAN-label switching (link local label): VLAN ID Label PBB-TE: Provider Backbone Bridges - Traffic engineering (domain wide label): DA_MAC + VLAN ID Label  Evolution of both Ethernet control and forwarding paradigm Legacy: VLAN Bridged Ethernet (MSTP) Provider bridges: IEEE 802.1q/.1ad PBB Provider Backbone Bridge: IEEE 802.1ah Shortest Path Bridging: IEEE 802.1aq (  link-state)  Evolution of Ethernet control paradigm only Shim header (sub-layer) Legacy: Ethernet Pseudo-wire over MPLS Packet Switched Network (PSN) Routing bridges (Rbridges) Ethernet packet-switched technology with two possible variants:  Ethernet Bridging: 802.1ah (PBB), 802.1aq (SPB)  Ethernet Switching (ongoing efforts):  MAC + VID based (domain-wide labels): 802.1Qay (PBB-TE)  VID based (link-local labels): Ethernet VLAN label switching

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Problem statement Management Spanning Tree, Learning, Filtering Forwarding PlaneEthernet Control (MSTP) Provisioning (Policy, etc) Provisioning (Forwarding Components) Existing IEEE forwarding components and their control does not fulfil requirements associated to Carrier Ethernet metro (and core) networks

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Evolution of control and forwarding paradigms: Ethernet VLAN-label Switching (ELS) Spanning Tree Protocol (STP) (VLAN-)Bridges Multiple STP (MSTP) Provider Bridges (PB) Multiple STP (MSTP) Provider Backbone Bridges (PBB) Ethernet Bridging branch (Distance vector) Ethernet Switching branch (Link State routing) S-VID (encapsulation) + Constraint-based switched data paths Link-local labels

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008  Ethernet LER (E-LER) function: take an incoming Ethernet MAC frame, add or remove the label (encoded in the TAG field)  Ethernet LSR (E-LSR): take incoming labelled Ethernet MAC frame and perform label swap (VID in  VID out) => forwarding independent of destination MAC address  Ethernet: point-to-point and point-to-multipoint data paths Ethernet VLAN-label Switching (ELS) - Overview Ethernet 802.1ad Switch E-LSR SourceDest Router S-VID swap S-VID push S-VID pop Ethernet LSP Ethernet MAC frame Eth. PHY Ethernet 802.1ad Switch Router PHY MAC header + S=VID Payload (Eth, X, Y)

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Ethernet VLAN-label Switching (ELS) - Framing

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Traffic engineering adapt traffic routing to network conditions with joint traffic and resource-oriented performance objectives  Effectively control usage of available network resources (put traffic where unused capacity is)  Efficiently re-/direct selected traffic flows from IGP shortest path onto an alternative path  Rapidly redistribute traffic in response to changes in network topology  Performance objectives (provisioning and recovery)  Resource-oriented  Traffic-oriented: packet loss, delay (and variation)  Approaches  Proactive (longer-term): anticipating traffic changes  Reactive/adaptive (shorter-term): responsive to traffic changes ELS Control Paradigm: Traffic Engineering

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Use Request Constraints Explicit Route Representation (GMPLS) RSVP-TE Signaling Traffic engineering Database (TEDB) Traffic engineering Database (TEDB) Routing table (GMPLS) OSPF-TE Operations performed by a LSP head-end (G)MPLS-TE capable node Constrained-SPF Computation (GMPLS) OSPF-TE Extensions Distributed (piggybacked) using Opaque Link State Advertisements (LSA) & encoded as Link sub-TLV Metrics: Unreserved Bandwidth, Maximum Reservable Bandwidth, TE Metric, Resource Class and ISCD (Max. LSP Bandwidth, Switching Cap., LSP Enc. Type) (1)Store information from IGP flooding in the Link State DB (LSDB) (2)Store traffic engineering information in the TE Link State DB (TEDB) (3)Examine user defined constraints for the incoming connectivity requests (=> QoS routing) (4)Path computation for the data path (LSP) through the TE link topology (=> Policy routing) (5)Representation of the computed path as an Explicit Route (=> Source routing) (6)Pass Explicit Route to (GMPLS) RSVP-TE engine for signaling Constrain-based Routing (Policy-based + QoS source routing)

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Ethernet VLAN-Label Switching (ELS) S-VID Label (link local) Positioning ELS Ethernet (Untagged, C-/S-VID) Shim (I-SID) Ethernet + B-VID Provider Backbone Bridges (PBB) Provider Backbone Bridges (PBB-TE) Ethernet (Untagged, C-VID) [Ethernet] + S-VID Ethernet (Untagged, C-/S-VID) Shim (CW + PW label) PSN Tunnel (MPLS) Ethernet PW over MPLS MPLS Label (link local) Payload IEEE: PBB/PBB-TE Ethernet VLAN-label Switching (ELS) Ethernet PW over MPLS 4k LSP per port (max.) LSP merging Unique payload type per LSP Encapsulating LSP can not be merged (as PW labels are node specific) PBB: same issues as for any other based technology PBB-TE: Single domain (MAC unicity) and no multicast support (single VID space segmentation)

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Positioning ELS vs PBB-TE (1) ELS (Ethernet VLAN-label Switching) Provider Backbone Bridges (PBB-TE) Paradigm Ethernet frame forwarding independent from destination MAC address (no learning) Constraint-based routing Add traffic engineering capabilities to PBB networks Connection ID encoded in the data frame Label encodingS-VID (12 bits)B-VID + B-MAC DA (requires MAC-in-MAC) Label semanticLink localDomain wide HierarchySingle level Path Unidirectional, Bi-directional P2P Unidirectional: MP2P (merge), P2MP (multicast) Unidirectional, Bi-directional P2P Unidirectional MP2P (multiplexing requires SA MAC lookup =/= classical label merging) No P2MP data path support Provisioning Recovery Control-plane based (GMPLS RSVP-TE) Management or optionally control-plane based (GMPLS RSVP-TE) Data plane linear protection based on (ongoing efforts on ring protection) Load balancingNo (in order delivery) OAM BFD, Ping, Traceroute ETH OAM (based on Y.1731) CC/CV OAM requires SA MAC lookup

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Router Frame Filtering I-SID B-VID B-DA IP/MPLS ETH + S-VID Router PWoMPLS LER IP/MPLS ETH+S-VID IP/MPLS ETH IP/MPLS ETH Ethernet LSR Ethernet Transport Positioning ELS vs PBB-TE (2) PB (BCB) VLAN label Switching Ethernet Label Switching S-VID Label Switching Frame forwarding independent of MAC address Same Ethernet MAC address space Disjoint Ethernet MAC address spaces ETH PHY PBB (BEB) IP/MPLS ETH + S-VID IP/MPLS ETH + S-VID B-SA

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Positioning ELS vs Ethernet PW over PSN (1) Ethernet Label Switching (ELS) Ethernet (Untagged, C-VID) [Ethernet] + S-VID Ethernet (Untagged, C-/S-VID) Shim (CW + PW label) PSN Tunnel (MPLS) Ethernet Pseudo-Wires (PW) Connectivity Service Network Emulation & Adaptation Network PE-to-PE Connection Data link layer Ethernet P2P, P2MP, MP Segment Pseudo-Wire (PW) label MPLS/T-MPLS Network Intermediate TrunksMPLS Tunnel/T-MPLS Tunnel Ethernet MAC/PPP-HDLC Ethernet PW over PSN Client PayloadOutside scope Physical layerEthernet PHY/SONET-SDH Ethernet Transport Ethernet P2P, P2MP Segment Ethernet Path (PE-to-PE) Append S-VID to Ethernet frames IP, IP/MPLS, etc. Ethernet PHY/SONET-SDH

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Router PWoMPLS LER IP/MPLS PW ETH MPLS DLL IP/MPLS ETH IP/MPLS ETH MPLS LSR Router PWoMPLS LER IP/MPLS ETH+S-VID IP/MPLS ETH IP/MPLS ETH Ethernet LSR Ethernet (connectivity) Service Ethernet Transport Positioning ELS vs Ethernet PW over PSN (2) MPLS Label Switching VLAN label Switching MPLS Label Switching Same Ethernet MAC address space Disjoint Ethernet MAC address spaces ETH PHY S-VID Label Switching Frame forwarding independent of MAC address

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Resolving the Ethernet Paradox Ethernet Paradox  Ethernet evolves as intra-domain aggregation technology for metro & core networks (by better adapting transport to Ethernet as MPLS is adapted to IP) Ethernet forwarding plane  Ethernet switching technology e.g. ELS  Moving Ethernet "networking" properties (linked to LAN / campus networks) toward metro-aggregation networks - but also core - definitely transform intrinsic nature of Ethernet Ethernet routing paradigm (control)  use of unified control e.g. GMPLS Consequences  Ethernet control:  From distance vector routing protocol (spanning tree protocol) to link state routing protocol  As IP routing evolved from RIP (distance vector) to OSPF (link state)  Ethernet forwarding:  Ethernet forwarding without specific mechanisms suitable/dedicated for LAN (campus, enterprise, etc.) environments  Mechanisms fitting specific needs of aggregation

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Router IP/MPLS Optical ETH IP/MPLS ETH IP/MPLS ETH Router IP/MPLS ETH+S-VID IP/MPLS ETH IP/MPLS ETH Ethernet LSR Ethernet (connectivity) between routers using OWS network ELS and Architectural evolution: IP over Optics  IP over Carrier Ethernet Optical Switching VLAN label Switching Optical Switching S-VID VLAN Label Switching (802.1ad) Same Ethernet MAC address space + Same admin domain Disjoint Ethernet MAC address spaces + Service boundary ETH PHY Ethernet (connectivity) between routers using carrier Ethernet switching network

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Architectural evolution: IP over Optics  IP over Carrier Ethernet  IP routers  traffic aggregation (level 1)  networking (single peering point), IP fast re-routing (not MPLS), and multi- topology routing, and BFD (OAM)  Carrier Ethernet: robust, resilient, flexible and cost-effective traffic aggregation (level 2)  Optical equipment/switching: (internal long distance) connectivity IP router Carrier Ethernet Optical Domain boundary Long distance, Ethernet switch interconnection Domain boundary IP router ETH PHY

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Tomorrow’s situation < 5,5 km< 50 km < 100 km< O(100 km) O(10) nodes Large CO Regional POP Ethernet aggregation Ethernet metro switch Core Metro-Aggregation Metro Access First Mile < 10 nodes IP edge routers Customer Premises IP Access router SP1..i-1 SPi…n Internet Ethernet aggregation Ethernet core switch 100GbE

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Conclusion: Evolution of Ethernet control and forwarding paradigms Forwarding component control Provisioning (TE data paths, re-routing, etc) Management Forwarding PlaneUnified Ethernet Control (e.g. GMPLS) The ultimate goal toward Carrier Ethernet … Provisioning (Forwarding Components)

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Several issues for further investigation Ethernet Forwarding Plane  Ethernet label space and scalability (  Label/LSP merging ?) - specific to link- local label switching based Ethernet forwarding  Ethernet CoS mechanisms (DSCP to Ethernet PCP mapping  DCP ?) - common  Ethernet multicast traffic (connectivity and adaptation) - common Ethernet Control - common  Unified traffic engineering (including fast re-routing)  lighter protocol suite(*) ?  Adaptive traffic engineering and resource allocation including Bandwidth Constraint Models (BCM)  Lightweight measurement/monitoring capabilities including performance (*) fundamental issue: developing, deploying and operating metro Ethernet using unified control must remain time-, resource- and cost-efficient (prevent over-engineering)

All Rights Reserved © Alcatel-Lucent | NGI Workshop | April 2008 Thanks ! Acknowledgements This work was carried out within the framework of the IWT TIGER project sponsored by the Flemish government institute for Innovation through Science and Technology in Flanders (IWT)