TRILL MPLS-Based Ethernet VPN Kingston Smiler, Mohammed Umair, Shaji Ravindranathan, Lucy Yong, Donald Eastlake draft-muks-trill-transport-over-mpls-01 draft-muks-trill-evpn-00 July 2015

Overview This Presentation covers providing Ethernet L2VPN service using TRILL. There are multiple L2VPN technologies already available e.g VPLS, BGP eVPN etc. VPLS is widely used in service provider networks, whereas BGP eVPN is widely used as a data center interconnect technology with MPLS / VxLAN tunnel. This draft introduces a new L2VPN technology using TRILL VTSD / VPTS. TRILL eVPN can be an alternate to both VPLS and BGP eVPN. In some case it provides more benefit, which neither of these technologies can meet, without any additional capability.

Key Terminologies and Introduction

VPTS model (Virtual Private TRILL Service) VPTS is similar to what VPLS does for a bridge domain. VPLS provides virtual private LAN service for different customers. This model provide Virtual Private TRILL service (VPTS) for different TRILL tenants. PE routers are replaced with TIR. A TIR is capable of running both the VPLS and TRILL Protocols. i.e (TIR = MPLS PE + TRILL RBridge). A new kind of switch domain, called VTSD (Virtual TRILL Switch Domain), replaces the VSI (bridge) in VPLS. The VTSD act as a Virtual RBridge inside TIR One VTSD per tenant / VPLS domain. The VTSD maintains all the information maintained by the RBridge for only that tenant. Forms adjacency with Other VTSD in the same VPLS domain. RBridge in the TRILL sites. VTSD should take care of segregating one customer traffic with other. No need to run Split Horizon in the provider PSN network, as TRILL takes care of Loop free topology using Distribution Trees. VLAN and MCast pruning of TRILL protocol take care of pruning the Distribution tree. July 2015 July 2015

VTSD Introduction According to “draft-muks-trill-transport-over-mpls-01”, a VTSD is logical RBridge inside TIR. It should be capable of performing all the operations a standard TRILL Switch should, one such Operation is the Appointed Forwarder mechanism. The other operation of Standard TRILL switch is support for multiple parallel links, Multipathing across such parallel connections can be freely done for unicast TRILL Data traffic on a per-flow basis but is restricted for multi-destination traffic.

VPLS based L2VPN service Site2 PE2 Site1 PE1 MPLS Cloud Site 3 VSI / VFI PE3 PWE3 mesh July 2015 July 2015

TRILL eVPN / VPTS TIR2 TIR1 MPLS Cloud TIR3 VTSD which is a R-Bridge Site2 TIR2 Site1 TIR1 MPLS Cloud Site 3 VTSD which is a R-Bridge TIR3 PWE3 mesh July 2015 July 2015

R-Bridge in VPTS Key Benefits. Brings in all the key benefits provided by TRILL to L2 networking into L2VPN network. Multihoming with all-active forwarding Multiple parallel paths Ring topology support in the Core

Parallel links PE2 PE1 MPLS Cloud PE3 VTSD which is a R-Bridge Site2 PE2 Site1 PE1 MPLS Cloud Site 3 VTSD which is a R-Bridge PE3 PWE3 mesh July 2015 July 2015

Ring Topology Support PE1 PE2 MPLS Cloud PE3 PE14 Site1 Site2 PE1 PE2 MPLS Cloud Site3 Site 4 PE3 PE14 VTSD which is a R-Bridge July 2015 July 2015

Use Case in Data Center with TOR to Core is L2 and MLAG MPLS / IP Network TOR Access Core / PE MLAG DataCenter 1 TOR Access Core / PE MLAG DataCenter 2 TOR Access Core / PE MLAG DataCenter 3

Datacenter Use Case Requirement Complete L2 Mesh inside the DC from ToR to Core Switch The DCI layer is connected as a RING with redundant link between the PEs (Core Switch) with MPLS Core. There is no MESH connectivity across the DCI layer. But there is a ring connectivity. Each Datacenter has two core switch to provide redundancy Each core switches in the DCI layer is connected to adjacent datacenter with 2 links, one link from each core. This provides a full redundancy and provide assured traffic flow until there are 4 or more link failures in the MPLS core.

BUM traffic flow MPLS / IP Network MLAG MLAG MLAG TOR TOR TOR Access Core / PE MLAG DataCenter 1 TOR Access Core / PE MLAG DataCenter 2 TOR Access Core / PE MLAG DataCenter 3

Unicast Traffic flow MPLS / IP Network MLAG MLAG MLAG TOR TOR TOR Access Core / PE MLAG DataCenter 1 TOR Access Core / PE MLAG DataCenter 2 TOR Access Core / PE MLAG DataCenter 3

Use Case in Datacenter with complete TRILL MPLS / IP Network TOR Access Core with VTSD DataCenter 1 DataCenter 2 DataCenter 3 TOR TO CORE and DCI is complete TRILL

L2VPN requirement from DCI network Multihoming with all-active forwarding TRILL support multiple active parallel links between the TRILL R-Bridges / traditional L2 bridges. For actively load balancing between Layer2 bridge and TRILL, TRILL can uses AF mechanism (VLAN based) TRILL Active-Active Access using draft-ietf-trill-centralized-replication and PseudoNickname (draft-ietf-trill-pseudonode-nickname) Effectively scaling the bandwidth by adding more links BUM traffic optimization TRILL forwards all the BUM traffic via the distribution tree. This provides a mechanism to avoid flooding of all BUM traffic across the L2VPN boundaries. These distribution trees can be pruned based on VLAN as well as multicast groups.

L2VPN requirement from DCI network IP/MAC information exchanging TRILL ESADI supports exchanging the MAC information. draft-ietf-trill-arp-optimization-01 proposes exchanging IP address along with MAC via esadi. Control over MAC learning TRILL ESADI can be enhanced to apply some rules/filter over exchanging the MAC address Efficient usage of mesh in the core network TRILL protocol by its nature, uses the mesh / parallel links efficiently. MAC mass-withdrawal TRILL ESADI can be enhanced to use MAC mass-withdrawal Always Use shortest Paths for Delivering Unicast Frames TRILL protocol by its nature, uses the shortest path.

VPLS vs BGP EVPN vs TRILL EVPN S.No Description VPLS BGP-eVPN TRILL eVPN 1 Multihoming with all-active forwarding No. Yes 2 Efficiently distributing the traffic across all-active links No Can be done by introducing pseudo-nick name and draft-ietf-trill-centralized-replication 3 Effectively scaling the bandwidth by adding more links (more than 2) in a multi-homing scenario Partial 4 Delivery of BUM traffic optimization Yes with Distribution tree and Pruning (VLAN and Multicast). 5 Autodiscovery of services Missing 6 Delivering Layer 2 and Layer 3 services over the same interface (Integrated Routing and Bridging) Can be done by enhancing esadi Two drafts already available draft-ietf-trill-arp-optimization-01 draft-ietf-trill-ia-appsubtlv-05 7 Multiple data plane encapsulation choices Will be addressed in next version of the draft. 8 Load-balancing inside the core network No efficient solutions 9 ARP/ND proxy We Have to make use of below two drafts Two drafts already available draft-ietf-trill-arp-optimization-01 draft-ietf-trill-ia-appsubtlv-05

VPLS vs BGP EVPN vs TRILL EVPN (Contd..) S.No Description VPLS BGP-eVPN TRILL eVPN 11 Control plane learning of MAC No Yes 12 Virtualization and isolation of different instances 13 Split Horizon 14 Mesh in the core network Not required 15 Efficient usage of links in the core network 16 BUM traffic handling Flood and learn Flood and learn / multicast tree Via distribution tree with VLAN pruning/Mcast Pruning 17 MAC mobility Need to check and propose a solution 18 MAC mass-withdrawal ESADI may need to be enhanced to do this. 19 Layer-3 ECMP – all links forwarding 20 Significantly larger Name-Space in the Overlay (16M segments) Yes TRILL with fine grained labelling (Stacking two vlan tags 24 bits, 2^24 (16M) entries) 21 Extensive OAM Capabilities 22 Frames can reach from one PE to other PE Via another PE NO N0 YES 23 Always Use shortest Paths for Delivering Unicast Frames There is no way to check 24 Can work in Ring topolgy without mesh / HUB and Spoke

Next Step Need to identify micro-level gaps in TRILL to provide efficient eVPN support. The drafts needs more work. Comments welcome.