Internet Draft Document Marc Lasserre Layer 2 VPN Working Group Xipeng Xiao draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Riverstone Networks Yetik Serbest Cesar Garrido SBC Telefonica Expires: August 2004 February 2004 VPLS Applicability draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. Abstract Virtual Private LAN Service (VPLS) is a layer 2 VPN service that provides multipoint connectivity in the form of an Ethernet emulated LAN, while usual L2 VPN services are typically point-to-point. Such Lasserre et al. [Page 1] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 emulated LANs can span metropolitan area networks as well as wide area networks. VPLS defines a method for signaling MPLS connections between member PEs of a VPN and a method for forwarding Ethernet frames over such connections. This document describes the applicability of such procedures to provide VPLS services. Conventions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 RELATED DOCUMENTS www.ietf.org/internet-drafts/draft-ietf-l2vpn-vpls-ldp-01.txt www.ietf.org/internet-drafts/draft-ietf-l3vpn-applicability- guidelines-00.txt Table of Contents Status of this Memo................................................1 Abstract...........................................................1 Conventions........................................................2 Intellectual Property Considerations...............................3 Full Copyright Statement...........................................3 1. VPLS Overview...................................................4 2. Operation of data, control and management planes................5 2.1. Control Plane.................................................5 2.1.1. Signaling...................................................5 2.2. Data Plane....................................................5 2.2.1. Ingress Processing..........................................5 2.2.2. Egress Processing...........................................5 2.2.3. Intermediate Node Processing................................6 2.3. Management plane..............................................6 2.3.1. VPLS OAM....................................................6 3. VPLS vs. alternative approaches.................................6 3.1. Ethernet switching............................................6 3.2. BGP VPN.......................................................7 4. Scalability.....................................................7 4.1. Mesh topology.................................................7 4.2. Signaling.....................................................7 4.3. MAC addresses and MAC learning................................7 4.4. Packet replication............................................7 Lasserre et al. [Page 2] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 4.5. Broadcast limiting............................................8 4.6. Multicast.....................................................8 5. Deployment Issues...............................................8 5.1. Provisioning..................................................8 5.1.1. VPLS membership management..................................8 5.1.2. Tunnel Provisioning.........................................9 5.2. PE Discovery..................................................9 5.3. Migration impacts............................................10 5.3.1. Existing L2 802.1Q VLAN Based Metro Infrastructure.........10 5.3.2. Existing IP Routed Environment.............................12 5.4. Multihoming..................................................12 5.5. Loop Prevention..............................................13 5.6. Packet re-ordering...........................................14 5.7. Multi-Domain VPLS Service....................................15 5.8. MTU (Maximum Transmission Unit) Issues.......................15 5.9. Interworking.................................................15 5.9.1. Interworking with BGP VPNs.................................15 5.9.2. Interworking with Frame Relay & ATM attachment circuits....15 5.10. Quality of Service..........................................15 5.11. Security....................................................16 5.11.1. Traffic Separation Between VPLS Instances.................16 5.11.2. Denial of Service (DoS)...................................16 6. Acknowledgments................................................17 7. References.....................................................17 8. Authors' Addresses.............................................18 Intellectual Property Considerations This document is being submitted for use in IETF standards discussions. Full Copyright Statement Copyright (C) The Internet Society (2001). All Rights Reserved. This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of Lasserre et al. [Page 3] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English. The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns. This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. 1. VPLS Overview The primary motivation behind Virtual Private LAN Services (VPLS) is to provide connectivity between geographically dispersed customer sites across MAN/WAN network(s), as if they were connected using a LAN. The intended applications for the end-user can be divided into the following two categories: - Connectivity between customer routers - Connectivity between customer Ethernet switches In addition, VPLS can also be used by the service provider to deliver services (e.g. triple play) to connected end-users. Unlike L3 VPNs such as BGP VPNs [BGP-VPN] where traffic exchanged between customers and service providers must be IP, VPLS only requires traffic to be Ethernet over which any protocol can be used, e.g. Netbios or IPX. The Service Provider Network is a packet switched network (PSN). The PEs are assumed to be fully meshed with transport tunnels over which customer frames that belong to a specific VPLS instance are encapsulated and forwarded. IP-in-IP, L2TPv3, GRE, and MPLS are examples of transport tunnels. Specific labels used to identify end-to-end paths over such tunnel LSPs are established via targeted LDP [VPLS-LDP]. These LSPs are known as pseudo-wires (PWs). VPLS defines the bridging rules required for PEs to provide an emulated Ethernet LAN service. In particular it defines how a loop- free topology must be built and the forwarding rules between PEs, along with the signaling method to set up PWs between PEs. Lasserre et al. [Page 4] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 The resulting service provides a unique broadcast domain per VPN, with the ability to send unicast, multicast and broadcast traffic (as well as flooding of unknown unicast traffic). 2. Operation of data, control and management planes 2.1. Control Plane 2.1.1. Signaling As with [PWE3-ETHERNET], [VPLS-LDP] specifies the use of targeted LDP for the signaling of PWs. PWs are established between PEs that are part of the same VPLS instance. 2.2. Data Plane 2.2.1. Ingress Processing VPLS provides an Ethernet emulated LAN service and hence customer frames are encapsulated as Ethernet frames (Ethernet DIX or 802.1). Note that such Ethernet frames can be carried over various access transport technologies (Frame Relay, ATM, etc). Ingress PEs will determine which Forwarding Information Base (FIB) to look up based on the port, VLAN or port/VLAN combination frames come from. This port to FIB mapping is performed at provisioning time. The destination MAC address is then looked up to determine on which PW this address has been learned from. If the lookup fails, i.e. if this MAC address has not been learned yet, the frame needs to be sent on all the PWs that are part of the corresponding VPLS instance. If the address is known, the frame is sent only over the associated PW. Before actually transmitting the customer frame, it needs to be encapsulated as defined in [PWE3-ETHERNET], and is further encapsulated with the appropriate transport header (e.g. MPLS or GRE). 2.2.2. Egress Processing Once the tunnel header has been removed, the egress PE determines from the PW label which FIB to look up to determine the egress port, VLAN or port/VLAN combination. The original Ethernet frame is then encapsulated with the proper transmission header if necessary (e.g. Frame Relay header) and sent over the corresponding port. MAC addresses are learned dynamically as traffic is exchanged. New source MAC addresses are learned on a per PW label per VPLS service instance basis. An aging timer is used to remove such bindings after a period of time. When user topology changes occur, MAC withdrawal messages in the signaling plane may be used to unlearn MAC addresses to improve convergence time. Lasserre et al. [Page 5] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 Egress PEs might also be configured to perform specific egress encapsulation functions (e.g. VLAN translation). 2.2.3. Intermediate Node Processing Intermediate nodes (P routers) only act as pure forwarders based on the outer tunnel header. Hence, they do not participate in any VPLS- related processing. Only PE routers maintain VPN specific information. This improves the scalability of VPLS service. 2.3. Management plane 2.3.1. VPLS OAM VPLS OAM is used to verify whether the VPLS instance of a particular customer works, and if not, where the fault is located. VPLS OAM is very important for the operations of VPLS networks. Currently there are two proposals for VPLS OAM. One proposal uses the OAM mechanisms being defined by the IEEE 802.1ad, 802.3ah, and ITU-T [Y.17ethoam] to verify MAC-layer connectivity status and locate fault at the PEs and MTUs. This approach is agnostic to PW type between the PEs or between PEs and the MTUs. The other proposal [STOKES] defines an MPLS-based approach to verify connectivity status and locate fault at the PEs and MTUs for VPLS deployments that use MPLS PWs. With VPLS OAM, ideally the OAM packets should always follow the same path as the VPLS data packets. However, because the Ethernet MAC layer has no TTL support, both approaches need to add something in the OAM packets to achieve the traceroute capability. As a result, neither approach can guarantee that traceroute packets always follow the same path as the VPLS data packets (without requiring change of existing network equipment). Therefore, both approaches are still evolving. Nevertheless, they achieve the practical purpose of verifying VPLS connectivity and locating fault to a good extent. 3. VPLS vs. alternative approaches 3.1. Ethernet switching Ethernet can be used to provide multipoint connectivity within small geographical areas such as small metropolitan networks. Pure Ethernet based solutions have scalability issues (e.g. STP limitations, 4095 VLAN limitations). Some enhancements such as QinQ, STP extensions (RSTP, MSTP) provide additional scalability. VPLS overcomes several of Ethernet based solutions by supporting large numbers of VPNs, better traffic engineering, and better quality of service. Lasserre et al. [Page 6] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 It is not uncommon for VPLS networks to be complemented with Ethernet switched networks as an aggregation layer. 3.2. BGP VPN In metropolitan area networks (MANs), BGP is usually not enabled. MANs provide a transport service to end-users. When multiple sites need to be connected within a metro, VPLS offers the appropriate multipoint transport solution. When multipoint connectivity is required across wide area networks such as national backbones, BGP VPNs can be more appropriate. Section 5.8.1 describes how VPLS and BGP VPNs can be complementary. 4. Scalability 4.1. Mesh topology A full mesh of tunnel LSPs, over which PWs are established – resulting in a full mesh of PWs, is created between participating PEs. When using hierarchical VPLS constructs, the size of this full mesh can be reduced to hub PEs aggregating point-to-point spokes as described in section 10 of [VPLS-LDP]. This reduces the number of tunnels and PWs from O(N*N) to O(N). 4.2. Signaling Using HVPLS constructs also allows the total number of targeted LDP sessions to be reduced from O(N*N) to O(N). 4.3. MAC addresses and MAC learning Depending on the type of CE devices used, i.e. switches or routers, the total number of MAC addresses to be learned by VPLS PEs can vary from one address per site to a large number of MAC addresses. When Ethernet networks exceed a large number of MAC addresses (e.g. hundreds), routers are introduced to limit the size of such broadcast domains. This reduces the total number of MAC addresses to learn to such routers only. In the case of large flat Ethernet networks, ingress PEs must be able to limit the number of MAC addresses that can be learned on a per VPLS basis. 4.4. Packet replication With VPLS, broadcast, multicast and unknown destination frames get replicated by the ingress PEs, i.e. close to the source of the Lasserre et al. [Page 7] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 frame. Ideally such frames should be replicated as close to the destination as possible to minimize bandwidth consumption. With hierarchical VPLS, the replication process is distributed between several ingress and egress MTUs and PEs. This helps not only minimizing bandwidth resources but also improving multicast performance and reducing latency. 4.5. Broadcast limiting Ingress MTUs or PEs may be able to rate limit the amount of broadcast traffic generated by end users in order to protect core resources and to prevent a few users from using all the bandwidth available. 4.6. Multicast In order to optimize the replication of multicast traffic, it is highly desirable for PEs to support multicast snooping techniques in order to only forward traffic where needed. In the case where the CE device is an L2 switch, IGMP snooping would be required, however, if the CE device is a router PIM snooping would be more applicable. 5. Deployment Issues 5.1. Provisioning 5.1.1. VPLS membership management As service providers start to provide Ethernet services to their customers, they build more complex end-to-end services for which the VPLS part plays a key role, but additional tools are also required: - Bandwidth control on customer facing ports - QoS classification and propagation - Traffic engineering per customer service - Activation of traffic monitoring and accounting mechanisms Managing the membership of customer sites to this service is not only keeping track of which PEs are part of a VPLS instance, but also the detailed characteristics of those connections, the path the customer traffic must use, the service profile it should have, etc. Auto-discovery mechanisms help to simplify the VPLS to PE membership management but are just a small part on the whole process for completely activating the end-to-end service. Therefore, it is highly convenient to make use of tools that can automate the provisioning tasks, in such a way that the service provider can take advantage, in a simplified way, of the traffic control mechanisms that MPLS provides, like Traffic Engineering. Lasserre et al. [Page 8] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 5.1.2. Tunnel Provisioning One of the key points for service providers delivering VPLS services is to take advantage of the capabilities provided by the LSPs signaled by RSVP. Examples of that are Traffic Engineering and Fast Restoration in case of failure, and being able to selectively use these capabilities on a per customer or service basis. Therefore, traffic engineered tunnels can become a service or customer differentiation mechanism, and not only an infrastructure communication path. Taking into account the importance of tunnel provisioning when activating VPLS services, it is highly desirable for service providers to be able to automate the creation and definition of the characteristics of the TE-LSPs, not only to take advantage of their properties but also to use the resources more efficiently, by creating LSPs only where needed and when needed. 5.2. PE Discovery PE auto-discovery is the process that a PE uses to discover IP addresses of other PEs in the network that participate in common VPLS instances as itself, so that these IP addresses needs not be configured by a network operator. Currently there are several proposals for PE auto-discovery: the BGP-based approach [VPLS-BGP], the RADIUS-based approach [RADIUS-DIS], and the Provisioning System- based approach. Because of the auto-discovery mechanism, at each PE, the network operator needs not specify other PEs' addresses. To add or remove a PE for a VPLS instance, the network operator may not need to touch other PEs (it still may be necessary to touch the other PEs in order to define customer specific service attributes such as per-PE QoS). As their names suggest, both approaches mandate the use of BGP or RADIUS in every PE, and rely on it to propagate the information of which PEs participate in a VPLS instance. The pros and cons of both approaches are discussed in their defining drafts. The key issues here are whether BGP should be in every VPLS PE and how suitable BGP is as a signaling protocol for VPLS. With the Provisioning System-based approach, network operators do not configure the PEs. Instead, they specify which PEs participate in which VPLS instances at the Provisioning System. The Provisioning System then translates such service information into PE configuration commands and telnet/ssh to the PEs to execute such commands. Because all information related to every VPLS instance is centralized at the Provisioning System, PE auto-discovery is automatically achieved. To add or remove a PE for a VPLS instance, a network operator simply specifies it at the Provisioning System Lasserre et al. [Page 9] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 which will then configure the PEs accordingly. For VPLS deployments that span across multiple domains, because the ASBRs (autonomous system border routers) of other domains can be treated as CEs of the current domain, these auto-discovery approaches can work in the inter-domain case as well. The scalability issues of such a scenario are discussed in [VPLS-BGP]. 5.3. Migration impacts Migration impacts may be mitigated through the use of careful planning when building and migrating the network. Also, consideration must be taken when integrating with protocols such as STP/MST and how control packets (BPDU’s) are handled. In addition, one must also consider ongoing standards efforts within various standards bodies such as the IEEE[802.1ad] and the Metro Ethernet Forum to assess future impact of any changes within the provider network. 5.3.1. Existing L2 802.1Q VLAN Based Metro Infrastructure 5.3.1.1. MPLS over 802.1Q(or QinQ) Tagged Infrastructure – Overlay Providers that have already deployed VLAN based architecture may choose to overlay an MPLS edge on top of this existing L2 domain. In this method, provider .1q tags maybe assigned to MPLS backbone links that are then used for carrying VPLS traffic. While this approach may allow for a simple transition to solve some immediate deficiencies of a pure L2 network, it still does not solve some of the underlying problems associated with protocols such as spanning tree. In this case, although MPLS may provide some scaling advantages, the limitations associated with spanning tree can still pose potential problems to the overall infrastructure. CE1 ------------------- ------ / / \ -|VPLS| / / \ / | PE |- / \ ------ / \ | 802.1Q/ | | QinQ | \ / ----- \ /\ ------ |VPLS|_/ \ / \ |VPLS| -| PE | \ / -| PE |- / ------ ------------------- ------ \ / \ \ CE3 --CE4 CE5 Lasserre et al. [Page 10] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 5.3.1.2. L2 Ethernet Islands Interconnected with a MPLS based VPLS core (HVPLS using Q or QinQ for spoke connections) Another mechanism that may be used for a migration strategy is to effectively utilize existing L2 (possibly .1Q based or QinQ) networks as “islands” attached to an MPLS based VPLS core network. In this particular case, the L2 network uses predetermined Provider .1Q tags (P-tags) to transport a given customers traffic. This P- tag is then utilized as a service delimiter that is then stripped prior to being transported across the MPLS cloud. The service delimiting P-tag is used to identify the VPLS instance to which the traffic should be mapped. A potential issue that can arise is the possibility of inadvertently creating an L2 loop in the event that the Ethernet access network(s) have redundant connections to the VPLS core. The assumption is that STP or another loop detection mechanism is already being utilized within the L2 domain and as such, should be utilized to perform loop avoidance when interconnecting with the VPLS core. ----CE1 ------- ------- / -------- CE2- / \ / PE1 / \ \ / \ / \ / \ ---| QinQ \ / MPLS/ \ / QinQ | | Domain PE VPLS PE Domain | \ / \ Domain / \ /\ \ / \ / \ / \ ------- ---------- -------- --CE3 Integration between VPLS and QinQ: A problem that may potentially arise when using VPLS to interconnect a traditional 802.1q access network to a QinQ access network revolves around the handling of .1q tags between the two access mechanisms. Customer connectivity at one site will be tied to a port on a VPLS PE/MTU that will utilize a PW for tunneling this packet through the network. Customer connectivity at another site will be interconnected to a port on a QinQ switch that will utilize QinQ techniques for transporting customer frames through the metro domain. The PE(s) responsible for interconnecting the MPLS domain to the QinQ island must perform additional operations to push or pop the QinQ Provider VLAN (P-VLAN) depending upon which direction the frame is being transported. In this particular case, on the VPLS egress facing the traditional CE, the PE must be capable of stripping the outer P-VLAN. On the VPLS egress facing the QinQ domain, the PE-rs must be capable of appending an additional P-tag prior to sending to the QinQ domain. Lasserre et al. [Page 11] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 ----- .-- Push P-VLAN / A1 \ ---- ----CE1 | / \ ------- ------- / | | | A2 CE2- / \ / PE1 \ / \ / \ / \ / \ ----- ---- ---| QinQ \ / MPLS/ | | Domain PE2 VPLS | \ / \ Domain / ----- \ / \ / |QinQ|_/ ------- ------- -| | Pop P-VLAN --. ---- / ------ ---- / \/ \ / \ CE = Customer Edge Router | A3 CE3 --C4 A4 | PE = Provider Edge Router \ / \ / ---- ---- 5.3.2. Existing IP Routed Environment Within an existing IP routed environment if the existing routers are capable of supporting MPLS, they may then be utilized as traditional P routers. If they are not MPLS capable, then alternate tunneling means such as GRE may be used. 5.4. Multihoming Multihoming is necessary in order to remove a VPLS PE as a single point of failure for all devices attached to it. There are two instances of multihoming that apply to VPLS: 1. When a CE device is connected to more than one PE, 2. In the case of hierarchical VPLS - when an MTU-s device is connected to more than one PE-rs. In both of these cases, the concern is that a particular MAC address will appear as a source on more than one PE device, causing other PE devices to continuously change their FIBs with regard to the true location of the MAC. This will cause constant table thrashing on the remote PEs, a behavior akin to a Layer 2 switch which participates in a loop. It is therefore required that any Layer 2 loops, created by multihoming of a CE or an MTU-s, be resolved within the group of devices participating in that loop. This group includes the multihomed CE or MTU-s, and all PEs to which it is attached. The PEs involved in such a loop are connected with a full mesh of pseudowires per VPLS instance. Lasserre et al. [Page 12] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 There are two approaches to resolving the loops created by the multihomed devices: 1. Running an MSTP instance between all devices in the group. In this case, the PEs within the group will need to utilize a P-VLAN for the purposes of running MSTP in the group. This P-VLAN can be re-used on non-overlapping groups of multihomed CE (or MTU-s) and its PEs. It must be clear that the MSTP process discussed here is a completely different and independent instance of STP than any STP the customer may be running. Such customer STP is always tunneled through the VPLS network, and is never acted upon by the PE or MTU-s devices. 2. The MTU-s or the CE can designate its link to one of the PEs it connects to as primary, and only send packets for this particular VPLS instance over that link. In this case the MTU-s (CE) is responsible for monitoring the state of that link and for switching to an alternate link if the primary fails. No action is required from the PEs participating in the group, though there should be an indication given from the MTU-s to its connected PEs as to whether the PE is connected to the primary or backup link. This is a very lightweight approach, which is quite useful given the simple and known topology between the CE (MTU-s) and its PEs. With this approach the operator must ensure that pseudowires in the core remain up, as long as the ingress PE they start from is up. This can typically be ensured with MPLS TE tools, such as fast re-route or back-up LSPs. If pseudowires in the core go down while their ingress PE is up and accepting customer traffic, blackholes can occur. In each case, the PE nodes are most likely in two different physical locations in the provider network providing network element protection, last mile protection, fiber diversity and provider facility backup. Customer STP traffic is always tunneled through the provider network, and is never acted upon by the PE or MTU-s devices. Lastly, it should be observed that, since VPLS services provide Ethernet switch-like transport level services, the customer is free to connect any device they desire as a CE. This could be anything from a simple host, hub, L2 switch, or a router. The operator has to be cognizant of the different capabilities of each of those devices to ensure loop-free environment when multi-homed. 5.5. Loop Prevention Loops in the core VPLS network are prevented by creating a full mesh of transport circuits between PEs and by applying a split-horizon rule. The split-horizon approach prevents a frame received from the Lasserre et al. [Page 13] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 backbone network from being sent out anything other than the customer facing ports belonging to that VPLS instance on the receiving PE. The frame MUST not be forwarded out other PW connecting the receiving PE to other PEs participating in the VPLS instance. This provides the necessary protection, network bandwidth optimization and scalability in the carriers network as it does not rely on link blocking technologies, like spanning tree type protocols. This forwarding mechanism allows PEs to effectively protect the core network from data loops. Customer networks need to be able to transparently transport the protocol information that allows their network to properly converge. However, the provider should consider loop protection schemes between the CE and PE that do not affect the customer functions. This would be in addition to spanning tree when the PE connects to a VLAN based L2 metro or when the customer is directly connected to multiple PE nodes. Methodologies providers can use to avoid loops when multi-homing CE devices have been discussed in the previous section. Some of these mechanisms involved running STP (or MSTP) between groups of PEs. The provider should look at deploying a loop protection scheme that would intervene automatically when it detects a loop condition. This loop protection scheme serves as an additional line of defence against protocol failures or misconfigurations, which can result in data loops. The concern is that a particular MAC address will appear as a source on more than one PE device, causing other PE devices to continuously update there tables. An external loop protection scheme adds a level of insurance above the customer link protection schemes. Its function is to reduce unnecessary core bandwidth usage when a loop condition occurs in an adjacent network and provide an extra level of protection to multihomed networks. It is a compliment but not a replacement for traditional loop protection mechanisms, like spanning tree. With directly connected customers, careful consideration needs to be given to backdoor connections. Backdoor connections provide an alternate path around a single provider. If a loop detection scheme is invoked here the customer may be forced to traverse a link that is not desired. 5.6. Packet re-ordering Normally there is only one transmission path towards a destination with VPLS so there is no packet re-ordering issue. But if some LSP load sharing mechanisms are used, packets may be re-ordered inside the PSN. If the users applications are sensitive to packet re- ordering, care must be taken to ensure packets are delivered in order. Lasserre et al. [Page 14] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 5.7. Multi-Domain VPLS Service As the use of VPLS grows, it is expected that customers will require a single VPLS service delivered by different providers (e.g. either for redundancy or physical location purposes). The different providers would then need to interconnect their VPLS domains for these customers. [VPLS-LDP] has provision for such a requirement, utilizing a single LSP tunnel between VPLS gateway devices. However, experience of such interconnection is not yet available. 5.8. MTU (Maximum Transmission Unit) Issues VPLS uses the same encapsulation of Ethernet frames that is defined in [PWE3-ETHERNET]. The MTU of transmission links used to transport [PWE3-ETHERNET] and VPLS traffic needs to accommodate the extra header used to carry the VC label and transport header. 5.9. Interworking 5.9.1. Interworking with BGP VPNs Typically when interworking VPLS with BGP VPNs, BGP VPNs are used to interconnect VPLS domains. In this type of scenario, BGP VPNs will be used to carry inter-metro (long-haul) traffic whereas intra- metro(local) traffic will be handled locally within the VPLS domain. Access/transport networks such as VPLS can be interconnected with BGP VPNs using various mechanisms such as Carriers-Carrier as defined in [RFC-2547]. A very common method for interconnection with BGP VPNs is to use a service delimiting tag (802.1Q VLAN-tag, VC-label, ATM VC, FR DLCI) to identify a customer’s traffic. This traffic is segregated and mapped to a given VRF using the delimiter. 5.9.2. Interworking with Frame Relay & ATM attachment circuits Frame Relay (FR) and ATM attachment circuits with Ethernet bridged encapsulation can be terminated within VPLS PEs. The resulting Ethernet frames (i.e. once the FR/ATM encapsulation has been stripped off) are processed as standard Ethernet frames. In order to support a complete interworking model between FR and Ethernet or between ATM and Ethernet, mapping service profiles and OAM traffic from one to the next will be necessary. Additionally, circuit management (e.g. LMI to PW state mapping) between the various technologies will be required. 5.10. Quality of Service Ingress PEs can classify incoming Ethernet traffic by either looking at 802.1P markings or by looking at L3 and/or L4 fields (e.g. Lasserre et al. [Page 15] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 ToS/DSCP field) contained within the payload to determine the frame class of service. The class of service determined during the classification phase can be mapped onto a corresponding class of service offered by the tunneling transport mechanism. For instance if MPLS tunneling is used, the appropriate EXP marking can be performed. Alternatively, the class of service can be mapped onto the appropriate tunnel which would have been explicitly traffic engineered to match the desired QoS. 5.11. Security 5.11.1. Traffic Separation Between VPLS Instances VPLS instances maintain separation of broadcast domains between themselves. Traffic entering a given VPLS instance at a given PE device does not, under any circumstances, cross the boundaries of the VPLS into another instance. VPLS devices (PEs and MTU-s) ensure that by maintaining a FIB table on a per-VPLS instance basis. The above statement is correct regardless of the learning mode employed by a particular VPLS instance (qualified or unqualified), or whether or not VLANs are treated as broadcast domain identifiers, or simply as circuit IDs which have no significance in determining the broadcast domain. In either of these cases, the VPLS instance is the outer-most "envelope" which ensures that traffic within it does not "leak" into another VPLS instance. 5.11.2. Denial of Service (DoS) Two types of DoS attacks are of concern with VPLS: 1. Attacks against VPLS devices 2. Attacks against other devices, for which the VPLS network is a transport. Attacks of the first type are naturally of greater concern for a VPLS operator, because they can destabilize the VPLS network as a whole, and affect multiple customers. The tunneling nature of VPLS by itself limits the possibilities for attacks via the data plane, simply because such attacks will be tunneled through the VPLS network, and will create the same load on the VPLS equipment as legitimate traffic will. Operators must watch for exception packet handling in VPLS equipment. In many cases, exception packets are sent to the control plane for handling. If that is the case, the operator must ensure that such exception packets can be rate-limited in a fashion that guarantees that the control plane will not be significantly burdened by them. Lasserre et al. [Page 16] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 The second type of DoS attacks, the ones that use the VPLS network as a transport, are not really a threat to the VPLS devices themselves, but are to devices behind them. VPLS PEs may be configured with rate-limiting and rate-shaping capabilities which permit them to limit the amount of traffic allowed into a particular VPLS instance. Optionally, they can also be tasked with advanced processing of the traffic they tunnel. For example, they may impose access lists which deny traffic from particular sources or protocols. Such approaches however are highly vendor-specific and outside the scope of [VPLS-LDP]. In addition, they may have significant design and operational repercussions. Alternative approaches which hand- off DoS protection activities to non-VPLS devices (such as customer equipment) are preferred. 6. Acknowledgments The authors wish to thank the following people for their constructive contributions to the text in this document: Javier Antich Ian Cowburn Richard Foote Rob Nath Nick Slabakov 7. References [802.1ad] "IEEE standard for Provider Bridges", Work in progress, December 2002. [PWE3-ETHERNET] "Encapsulation Methods for Transport of Ethernet Frames Over IP/MPLS Networks", draft-ietf-pwe3-ethernet-encap- 02.txt, Work in progress, February 2003. [PWE3-CTRL] "Transport of Layer 2 Frames Over MPLS", draft-ietf- pwe3-control-protocol-02.txt, Work in progress, February 2003. [L2FRAME] "L2VPN Framework", draft-ietf-ppvpn-l2-framework-03, Work in progress, February 2003. [L2VPN-REQ] "Service Requirements for Layer 2 Provider Provisioned Virtual Private Networks", draft-ietf-ppvpn-l2vpn-requirements- 00.txt, Work in progress, May 2003. [RADIUS-DIS] "Using Radius for PE-Based VPN Discovery", Work in progress, Jun. 2003 Lasserre et al. [Page 17] ID draft-lasserre-l2vpn-vpls-ldp-applic-00.txt Mar 2004 [STOKES] "Testing Hierarchical Virtual Private LAN Services", Work in progress, Jun. 2003 [VPLS-LDP] "Virtual Private LAN Services over MPLS", draft-ietf- ppvpn-vpls-ldp-01.txt, Work in progress, November 2003 [VPLS-BGP] "Virtual Private LAN Service", draft-ietf-ppvpn-vpls-bgp- 01.txt, Work in progress [Y.17ethoam] "OAM mechanisms for Ethernet based networks", ITU-T, SG13, Jul. 2003 8. Authors' Addresses Marc Lasserre Riverstone Networks Email: marc@riverstonenet.com Xipeng Xiao Riverstone Networks Email: xxiao@riverstonenet.com Yetik Serbest SBC Communications serbest@tri.sbc.com Cesar Garrido, Telefonica cesar.garridosanahuja@telefonica.es Lasserre et al. [Page 18]