3.2.4.1. MAC Learning Protection

In a Layer 2 environment, subscribers or customers connected to SAPs A or B can create a denial of service attack by sending packets sourcing the gateway MAC address. This will move the learned gateway MAC from the uplink SDP/SAP to the subscriber’s or customer’s SAP causing all communication to the gateway to be disrupted. If local content is attached to the same VPLS (D), a similar attack can be launched against it. Communication between subscribers or customers is also disallowed but split horizon will not be sufficient in the topology shown in Figure 60.

Figure 60:  MAC Learning Protection 

The 7450 ESS, 7750 SR, and 7950 XRS routers enable MAC learning protection capability for SAPs and SDPs. With this mechanism, forwarding and learning rules apply to the non-protected SAPs. Assume hosts H1, H2, and H3 (Figure 60) are non-protected while IES interfaces G and H are protected. When a frame arrives at a protected SAP/SDP, the MAC is learned as usual. When a frame arrives from a non-protected SAP or SDP, the frame must be dropped if the source MAC address is protected and the MAC address is not relearned. The system allows only packets with a protected MAC destination address.

The system can be configured statically. The addresses of all protected MACs are configured. Only the IP address can be included and use a dynamic mechanism to resolve the MAC address (cpe-ping). All protected MACs in all VPLS instances in the network must be configured.

To eliminate the ability of a subscriber or customer to cause a DoS attack, the node restricts the learning of protected MAC addresses based on a statically defined list. Also, the destination MAC address is checked against the protected MAC list to verify that a packet entering a restricted SAP has a protected MAC as a destination.

3.2.4.2. DEI in IEEE 802.1ad

The IEEE 802.1ad-2005 standard allows drop eligibility to be conveyed separately from priority in Service VLAN TAGs (S-TAGs) so that all of the previously introduced traffic types can be marked as drop eligible. The S-TAG has a new format where the priority and discard eligibility parameters are conveyed in the 3-bit Priority Code Point (PCP) field and, respectively, in the DE bit (Figure 61).

Figure 61:  DE Bit in the 802.1ad S-TAG 

The DE bit allows the S-TAG to convey eight forwarding classes/distinct emission priorities, each with a drop eligible indication.

When the DE bit is set to 0 (DE=FALSE), the related packet is not discarded eligible. This is the case for the packets that are within the CIR limits and must be given priority in case of congestion. If the DEI is not used or backwards compliance is required, the DE bit should be set to zero on transmission and ignored on reception.

When the DE bit is set to 1 (DE=TRUE), the related packet is discarded eligible. This is the case for the packets that are sent above the CIR limit (but below the PIR). In case of congestion, these packets will be the first ones to be dropped.

3.2.7.1. Selective MAC Address Learning

Source MAC addresses are learned in a VPLS service by default with an entry allocated in the FDB for each address on all line cards. Therefore, all MAC addresses are considered to be global. This operation can be modified so that the line card allocation of some MAC addresses is selective, based on where the service has a configured object.

An example of the advantage of selective MAC address learning is for services to benefit from the higher MAC address scale of some line cards (particularly for network interfaces used by mesh or spoke-SDPs, EVPN-VXLAN tunnels, and EVPN-MPLS destinations) while using lower MAC address scale cards for the SAPs.

Selective MAC addresses are those learned locally and dynamically in the data path (displayed in the show output with type “L”) or by EVPN (displayed in the show output with type “Evpn”, excluding those with the sticky bit set, which are displayed with type “EvpnS”). An exception is when a MAC address configured as a conditional static MAC address is learned dynamically on an object other than its monitored object; this can be displayed with type “L” or “Evpn” but is learned as global because of the conditional static MAC configuration.

Selective MAC addresses have FDB entries allocated on line cards where the service has a configured object. When a MAC address is learned, it is allocated an FDB entry on all line cards on which the service has a SAP configured (for LAG or Ethernet tunnel SAPs, the MAC address is allocated an FDB entry on all line cards on which that LAG or Ethernet tunnel has configured ports) and on all line cards that have a network interface port if the service is configured with VXLAN, EVPN-MPLS, or a mesh or spoke-SDP.

When using selective learning in an I-VPLS service, the learned C-MACs are allocated FDB entries on all the line cards where the I-VPLS service has a configured object and on the line cards on which the associated B-VPLS has a configured object. When using selective learning in a VPLS service with allow-ip-intf-bind configured (for it to become an R-VPLS), FDB entries are allocated on all line cards on which there is an IES or VPRN interface.

If a new configured object is added to a service and there are sufficient MAC FDB resources available on the new line cards, the selective MAC addresses present in the service are allocated on the new line cards. Otherwise, if any of the selective MAC addresses currently learned in the service cannot be allocated an FDB entry on the new line cards, those MAC addresses will be deleted from all line cards. Such a deletion increments the FailedMacCmplxMapUpdts statistic displayed in the tools dump service vpls-fdb-stats output.

When the set of configured objects changes for a service using selective learning, the system must reallocate its FDB entries accordingly, which can cause FDB entry “allocate” or “free” operations to become pending temporarily. The pending operations can be displayed using the tools dump service id fdb command.

When a global MAC address is to be learned, there must be a free FDB entry in the service and system FDBs and on all line cards in the system for it to be accepted. When a selective MAC address is to be learned, there must be a free FDB entry in the service and system FDBs and on all line cards where the service has a configured object for it to be accepted.

To demonstrate the selective MAC address learning logic, consider the following:

This is shown in Table 14.

Figure 62 shows the FDB entry allocation when the MAC addresses are global and when they are selective. Notice that in the selective case, all MAC addresses are allocated FDB entries on line card 2, but line card 1 and 3 only have FDB entries allocated for services VPLS 1 and VPLS 2, respectively.

Figure 62:  MAC FDB Entry Allocation: Global versus Selective  

Selective MAC address learning can be enabled as follows within any VPLS service, except for B-VPLS and R-VPLS services:

Enabling selective MAC address learning has no effect on single line card systems.

When selective learning is enabled or disabled in a VPLS service, the system may need to reallocate FDB entries; this can cause temporary pending FDB entry allocate or free operations. The pending operations can be displayed using the tools dump service id fdb command.

3.2.7.2. System FDB Size

The system FDB table size is configurable as follows:

where table-size can have values in the range from 255999 to 2047999 (2000k).

The default, minimum, and maximum values for the table size are dependent on the chassis type. To support more than 500k MAC addresses, the CPMs provisioned in the system must have at least 16 GB memory. The maximum system FDB table size also limits the maximum FDB table size of any card within the system.

The actual achievable maximum number of MAC addresses depends on the MAC address scale supported by the active cards and whether selective learning is enabled.

If an attempt is made to configure the system FDB table size such that:

The logic when attempting a rollback is similar; however, when rolling back to a configuration where the system FDB table size is smaller than the current system FDB table size, the system will flush all learned MAC addresses (by performing a shutdown then no shutdown in all VPLS services) to allow the rollback to continue.

The system FDB table size can be larger than some of the line card FDB sizes, resulting in the possibility that the current number of allocated global MAC addresses is larger than the maximum FDB size supported on some line cards. When a new line card is provisioned, the system checks whether the line card's FDB can accommodate all of the currently allocated global MAC addresses. If it can, then the provisioning succeeds; if it cannot, then the provisioning fails and an error is reported. If the provisioning fails, the number of global MACs allocated must be reduced in the system to a number that the new line card can accommodate, then the card-type must be reprovisioned.

3.2.7.3. Per-VPLS Service FDB Size

The following MAC table management features are available for each instance of a SAP or spoke-SDP within a particular VPLS service instance:

3.2.7.4. System FDB Size Alarms

High and low watermark alarms give warning when the system MAC FDB usage is high. An alarm is generated when the number of FDB entries allocated in the system FDB reaches 95% of the total system FDB table size and is cleared when it reduces to 90% of the system FDB table size. These percentages are not configurable.

3.2.7.5. Line Card FDB Size Alarms

High and low watermark alarms give warning when a line card's MAC FDB usage is high. An alarm is generated when the number of FDB entries allocated in a line card FDB reaches 95% of its maximum FDB table size and is cleared when it reduces to 90% of its maximum FDB table size. These percentages are not configurable.

3.2.7.6. Per VPLS FDB Size Alarms

The size of the VPLS FDB can be configured with a low watermark and a high watermark, expressed as a percentage of the total FDB size limit. If the actual FDB size grows above the configured high watermark percentage, an alarm is generated. If the FDB size falls below the configured low watermark percentage, the alarm is cleared by the system.

3.2.7.7. Local and Remote Aging Timers

Like a Layer 2 switch, learned MACs within a VPLS instance can be aged out if no packets are sourced from the MAC address for a specified period of time (the aging time). In each VPLS service instance, there are independent aging timers for locally learned MAC and remotely learned MAC entries in the FDB. A local MAC address is a MAC address associated with a SAP because it ingressed on a SAP. A remote MAC address is a MAC address received by an SDP from another router for the VPLS instance. The local-age timer for the VPLS instance specifies the aging time for locally learned MAC addresses, and the remote-age timer specifies the aging time for remotely learned MAC addresses.

In general, the remote-age timer is set to a longer period than the local-age timer to reduce the amount of flooding required for unknown destination MAC addresses. The aging mechanism is considered a low priority process. In most situations, the aging out of MAC addresses happens within tens of seconds beyond the age time. However, it, can take up to two times their respective age timer to be aged out.

3.2.7.8. Disable MAC Aging

The MAC aging timers can be disabled, which will prevent any learned MAC entries from being aged out of the FDB. When aging is disabled, it is still possible to manually delete or flush learned MAC entries. Aging can be disabled for learned MAC addresses on a SAP or a spoke-SDP of a VPLS service instance.

3.2.7.9. Disable MAC Learning

When MAC learning is disabled for a service, new source MAC addresses are not entered in the VPLS FDB, whether the MAC address is local or remote. MAC learning can be disabled for individual SAPs or spoke-SDPs.

3.2.7.10. Unknown MAC Discard

Unknown MAC discard is a feature that discards all packets ingressing the service where the destination MAC address is not in the FDB. The normal behavior is to flood these packets to all endpoints in the service.

Unknown MAC discard can be used with the disable MAC learning and disable MAC aging options to create a fixed set of MAC addresses allowed to ingress and traverse the service.

3.2.7.11. VPLS and Rate Limiting

Traffic that is normally flooded throughout the VPLS can be rate limited on SAP ingress through the use of service ingress QoS policies. In a service ingress QoS policy, individual queues can be defined per forwarding class to provide shaping of broadcast traffic, MAC multicast traffic, and unknown destination MAC traffic.

3.2.7.12. MAC Move

The MAC move feature is useful to protect against undetected loops in a VPLS topology as well as the presence of duplicate MACs in a VPLS service.

If two clients in the VPLS have the same MAC address, the VPLS will experience a high relearn rate for the MAC. When MAC move is enabled, the 7450 ESS, 7750 SR, or 7950 XRS will shut down the SAP or spoke-SDP and create an alarm event when the threshold is exceeded.

MAC move allows sequential order port blocking. By configuration, some VPLS ports can be configured as “non-blockable”, which allows a simple level of control of which ports are being blocked during loop occurrence. There are two sophisticated control mechanisms that allow blocking of ports in a sequential order:

For the first control mechanism, configuration CLI is extended by definition of “primary” and “secondary” ports. Per default, all VPLS ports are considered “tertiary” ports unless they are explicitly declared primary or secondary. The order of blocking will always follow a strict order starting from tertiary to secondary, and then primary.

The definition of criteria for the second control mechanism is the number of periods during which the specified relearn rate has been exceeded. The mechanism is based on the cumulative factor for every group of ports. Tertiary VPLS ports are blocked if the relearn rate exceeds the configured threshold during one period, while secondary ports are blocked only when relearn rates are exceeded during two consecutive periods, and primary ports when exceeded during three consecutive periods. The retry timeout period must be larger than the period before blocking the highest priority port so that the retry timeout sufficiently spans across the period required to block all ports in sequence. The period before blocking the highest priority port is the cumulative factor of the highest configured port multiplied by 5 seconds (the retry timeout can be configured through the CLI).

3.2.7.13. Auto-Learn MAC Protect

This section provides information about auto-learn-mac-protect and restrict-protected-src discard-frame features.

VPLS solutions usually involve learning MAC addresses in order for traffic to be forwarded to the correct SAP/SDP. If a MAC address is learned on the wrong SAP/SDP, traffic would be redirected away from its intended destination. This could occur through a misconfiguration, a problem in the network, or by a malicious source creating a DoS attack, and is applicable to any type of VPLS network; for example, mobile backhaul or residential service delivery networks. The auto-learn-mac-protect feature can be used to safeguard against the possibility of MAC addresses being learned on the wrong SAP/SDP.

This feature provides the ability to automatically protect source MAC addresses that have been learned on a SAP or a spoke/mesh SDP and prevent frames with the same protected source MAC address from entering into a different SAP/spoke or mesh SDP.

This is a complementary solution to features such as mac-move and mac-pinning, but has the advantage that MAC moves are not seen and it has a low operational complexity. If a MAC is initially learned on the wrong SAP/SDP, the operator can clear the MAC from the MAC FDB in order for it to be relearned on the correct SAP/SDP.

Two separate commands are used, which provide the configuration flexibility of separating the identification (learning) function from the application of the restriction (discard).

The auto-learn-mac-protect and restrict-protected-src commands allow the following functions:

If the auto-learn-mac-protect or restrict-protected-src discard-frame feature is configured under an SHG, the operation applies only to SAPs in the SHG, not to spoke-SDPs in the SHG. If required, these parameters can also be enabled explicitly under specific SAPs/spoke-SDPs within the SHG.

Applying or removing auto-learn-mac-protect or restrict-protected-src discard-frame to/from a SAP, spoke or mesh SDP, or SHG, will clear the MACs on the related objects (for the SHG, this results in clearing the MACs only on the SAPs within the SHG).

The use of restrict-protected-src discard-frame is mutually exclusive with both the restrict-protected-src [alarm-only] command and with the configuration of manually protected MAC addresses, using the mac-protect command, within a specified VPLS.

The following rules govern the changes to the state of protected MACs:

If a MAC address does legitimately move between SAPs/spoke or mesh SDPs after it has been automatically protected on a specified SAP/spoke or mesh SDP (thereby causing discards when received on the new SAP/spoke or mesh SDP), the operator must manually clear the MAC from the FDB for it to be learned in the new/correct location.

MAC addresses that are manually created (using static-mac, static-host with a MAC address specified, or oam mac-populate) will not be protected even if they are configured on a SAP/spoke or mesh SDP that has auto-learn-mac-protect enabled on it. Also, the MAC address associated with an R-VPLS IP interface is protected within its VPLS service such that frames received with this MAC address as the source address are discarded (this is not based on the auto-learn MAC protect function). However, VRRP MAC addresses associated with an R-VPLS IP interface are not protected either in this way or using the auto-learn MAC protect function.

MAC addresses that are dynamically created (learned, using static-host with no MAC address specified, or lease-populate) will be protected when the MAC address is learned on a SAP/spoke or mesh SDP that has auto-learn-mac-protect enabled on it.

The actions of the following features are performed in the order listed.

3.2.7.13.1. Operation

Figure 63 shows a specific configuration using auto-learn-mac-protect and restrict-protected-src discard-frame in order to describe their operation for the 7750 SR, 7450 ESS, or 7950 XRS.

Figure 63:  Auto-Learn-Mac-Protect Operation 

A VPLS service is configured with SAP1 and SDP1 connecting to access devices and SAP2, SAP3, and SDP2 connecting to the core of the network. The auto-learn-mac-protect feature is enabled on SAP1, SAP3, and SDP1, and restrict-protected-src discard-frame is enabled on SAP1, SDP1, and SDP2. The following series of events describes the details of the functionality:

Assume that the FDB is empty at the start of each sequence.

Sequence 1:

1. A frame with source MAC A enters SAP1, MAC A is learned on SAP1, and MAC-A/SAP1 is protected because of the presence of the auto-learn-mac-protect on SAP1.
2. All subsequent frames with source MAC A entering SAP1 are forwarded into the VPLS.
3. Frames with source MAC A enter either SDP1 or SDP2, these frames are discarded, and an alarm indicating MAC A and SDP1/SDP2 is initiated because of the presence of the restrict-protected-src discard-frame on SDP1/SDP2.
4. The above continues, with MAC-A/SAP1 protected in the FDB until MAC A on SAP1 is removed from the FDB.

Sequence 2:

1. A frame with source MAC A enters SAP1, MAC A is learned on SAP1, and MAC-A/SAP1 is protected because of the presence of the auto-learn-mac-protect on SAP1.
2. A frame with source MAC A enters SAP2. As restrict-protected-src is not enabled on SAP2, MAC A is relearned on SAP2 (but not protected), replacing the MAC-A/SAP1 entry in the FDB.
3. All subsequent frames with source MAC A entering SAP2 are forwarded into the VPLS. This is because restrict-protected-src is not enabled SAP2 and auto-learn-mac-protect is not enabled on SAP2, so the FDB is not changed.
4. A frame with source MAC A enters SAP1, MAC A is relearned on SAP1, and MAC-A/SAP1 is protected because of the presence of the auto-learn-mac-protect on SAP1.

Sequence 3:

1. A frame with source MAC A enters SDP2, MAC A is learned on SDP2, but is not protected as auto-learn-mac-protect is not enabled on SDP2.
2. A frame with source MAC A enters SDP1, and MAC A is relearned on SDP1 because previously it was not protected. Consequently, MAC-A/SDP1 is protected because of the presence of the auto-learn-mac-protect on SDP1.

Sequence 4:

1. A frame with source MAC A enters SAP1, MAC A is learned on SAP1, and MAC-A/SAP1 is protected because of the presence of the auto-learn-mac-protect on SAP1.
2. A frame with source MAC A enters SAP3. As restrict-protected-src is not enabled on SAP3, MAC A is relearned on SAP3 and the MAC-A/SAP1 entry is removed from the FDB with MAC-A/SAP3 being added as protected to the FDB (because auto-learn-mac-protect is enabled on SAP3).
3. All subsequent frames with source MAC A entering SAP3 are forwarded into the VPLS.
4. A frame with source MAC A enters SAP1, these frames are discarded, and an alarm indicating MAC A and SAP1 is initiated because of the presence of the restrict-protected-src discard-frame on SAP1.

Example Use

Figure 64 shows a possible configuration using auto-learn-mac-protect and restrict-protected-src discard-frame in a mobile backhaul network, with the focus on PE1 for the 7750 SR or 7950 XRS.

Figure 64:  Auto-Learn-Mac-Protect Example 

To protect the MAC addresses of the BNG/RNCs on PE1, the auto-learn-mac-protect command is enabled on the pseudowires connecting PE1 to PE2 and PE3. Enabling the restrict-protected-src discard-frame command on the SAPs toward the eNodeBs will prevent frames with the source MAC addresses of the BNG/RNCs from entering PE1 from the eNodeBs.

The MAC addresses of the eNodeBs are protected in two ways. In addition to the above commands, enabling the auto-learn-mac-protect command on the SAPs toward the eNodeBs will prevent the MAC addresses of the eNodeBs being learned on the wrong eNodeB SAP. Enabling the restrict-protected-src discard-frame command on the pseudowires connecting PE1 to PE2 and PE3 will protect the eNodeB MAC addresses from being learned on the pseudowires. This may happen if their MAC addresses are incorrectly injected into VPLS 40 on PE2/PE3 from another eNodeB aggregation PE.

The above configuration is equally applicable to other Layer 2 VPLS-based aggregation networks; for example, to business or residential service networks.

3.2.9.1. Spanning Tree Operating Modes

Per VPLS instance, a preferred STP variant can be configured. The STP variants supported are:

While the 7450 ESS, 7750 SR, or 7950 XRS initially use the mode configured for the VPLS, it will dynamically fall back (on a per-SAP basis) to STP (IEEE 802.1D-1998) based on the detection of a BPDU of a different format. A trap or log entry is generated for every change in spanning tree variant.

Some older 802.1w compliant RSTP implementations may have problems with some of the features added in the 802.1D-2004 standard. Interworking with these older systems is improved with the comp-dot1w mode. The differences between the RSTP mode and the comp-dot1w mode are:

The 7450 ESS, 7750 SR, and 7950 XRS support two BPDU encapsulation formats, and can dynamically switch between the following supported formats (on a per-SAP basis):

3.2.9.2. Multiple Spanning Tree

The Multiple Spanning Tree Protocol (MSTP) extends the concept of IEEE 802.1w RSTP by allowing grouping and associating VLANs to Multiple Spanning Tree Instances (MSTI). Each MSTI can have its own topology, which provides architecture enabling load balancing by providing multiple forwarding paths. At the same time, the number of STP instances running in the network is significantly reduced as compared to Per VLAN STP (PVST) mode of operation. Network fault tolerance is also improved because a failure in one instance (forwarding path) does not affect other instances.

The Nokia implementation of M-VPLS is used to group different VPLS instances under a single RSTP instance. Introducing MSTP into the M-VPLS allows interoperating with traditional Layer 2 switches in an access network and provides an effective solution for dual homing of many business Layer 2 VPNs into a provider network.

3.2.9.3. MSTP for QinQ SAPs

MSTP runs in a M-VPLS context and can control SAPs from source VPLS instances. QinQ SAPs are supported. The outer tag is considered by MSTP as part of VLAN range control.

3.2.9.4. Provider MSTP

Provider MSTP is specified in IEEE-802.1ad-2005. It uses a provider bridge group address instead of a regular bridge group address used by STP, RSTP, and MSTP BPDUs. This allows for implicit separation of source and provider control planes.

The 802.1ad access network sends PBB PE P-MSTP BPDUs using the specified MAC address and also works over QinQ interfaces. P-MSTP mode is used in PBBN for core resiliency and loop avoidance.

Similar to regular MSTP, the STP mode (for example, PMSTP) is only supported in VPLS services where the m-VPLS flag is configured.

3.2.9.5. Enhancements to the Spanning Tree Protocol

To interconnect PE devices across the backbone, service tunnels (SDPs) are used. These service tunnels are shared among multiple VPLS instances. The Nokia implementation of the STP incorporates some enhancements to make the operational characteristics of VPLS more effective. The implementation of STP on the router is modified in order to guarantee that service tunnels will not be blocked in any circumstance without imposing artificial restrictions on the placement of the root bridge within the network. The modifications introduced are fully compliant with the 802.1D-2004 STP specification.

When running MSTP, spoke-SDPs cannot be configured. Also, ensure that all bridges connected by mesh SDPs are in the same region. If not, the mesh will be prevented from becoming active (trap is generated).

To achieve this, all mesh SDPs are dynamically configured as either root ports or designated ports. The PE devices participating in each VPLS mesh determine (using the root path cost learned as part of the normal protocol exchange) which of the 7450 ESS, 7750 SR, or 7950 XRS devices is closest to the root of the network. This PE device is internally designated as the primary bridge for the VPLS mesh. As a result of this, all network ports on the primary bridges are assigned the designated port role and therefore remain in the forwarding state.

The second part of the solution ensures that the remaining PE devices participating in the STP instance see the SDP ports as a lower-cost path to the root rather than a path that is external to the mesh. Internal to the PE nodes participating in the mesh, the SDPs are treated as zero-cost paths toward the primary bridge. As a consequence, the path through the mesh is seen as lower cost than any alternative and the PE node will designate the network port as the root port. This approach ensures that network ports always remain in forwarding state.

In combination, these two features ensure that network ports will never be blocked and will maintain interoperability with bridges external to the mesh that are running STP instances.

3.2.10.1. Spoke SDP Redundancy for Metro Interconnection

When two or more meshed VPLS instances are interconnected by redundant spoke-SDPs (as shown in Figure 66), a loop in the topology results. To remove such a loop from the topology, STP can be run over the SDPs (links) that form the loop, such that one of the SDPs is blocked. As running STP in each and every VPLS in this topology is not efficient, the node includes functionality that can associate a number of VPLSs to a single STP instance running over the redundant SDPs. Therefore, node redundancy is achieved by running STP in one VPLS and applying the conclusions of this STP to the other VPLS services. The VPLS instance running STP is referred to as the “management VPLS” or M-VPLS.

If the active node fails, STP on the management VPLS in the standby node will change the link states from disabled to active. The standby node will then broadcast a MAC flush LDP control message in each of the protected VPLS instances, so that the address of the newly active node can be relearned by all PEs in the VPLS.

It is possible to configure two management VPLS services, where both VPLS services have different active spokes (this is achieved by changing the path cost in STP). By associating different user VPLSs with the two management VPLS services, load balancing across the spokes can be achieved.

Figure 66:  HVPLS with Spoke Redundancy 

3.2.10.2. Spoke SDP Based Redundant Access

This feature provides the ability to have a node deployed as MTUs (Multi-Tenant Units) to be multi-homed for VPLS to multiple routers deployed as PEs without requiring the use of M-VPLS.

In the configuration example shown in Figure 66, the MTUs have spoke-SDPs to two PE devices. One is designated as the primary and one as the secondary spoke-SDP. This is based on a precedence value associated with each spoke.

The secondary spoke is in a blocking state (both on receive and transmit) as long as the primary spoke is available. When the primary spoke becomes unavailable (due to link failure, PEs failure, and so on), the MTUs immediately switch traffic to the backup spoke and start receiving traffic from the standby spoke. Optional revertive operation (with configurable switch-back delay) is supported. Forced manual switchover is also supported.

To speed up the convergence time during a switchover, MAC flush is configured. The MTUs generate a MAC flush message over the newly unblocked spoke when a spoke change occurs. As a result, the PEs receiving the MAC flush will flush all MACs associated with the impacted VPLS service instance and forward the MAC flush to the other PEs in the VPLS network if “propagate-mac-flush” is enabled.

3.2.10.3. Inter-Domain VPLS Resiliency Using Multi-Chassis Endpoints

Inter-domain VPLS refers to a VPLS deployment where sites may be located in different domains. An example of inter-domain deployment can be where different metro domains are interconnected over a Wide Area Network (Metro1-WAN-Metro2) or where sites are located in different autonomous systems (AS1-ASBRs-AS2).

Multi-chassis endpoint (MC-EP) provides an alternate solution that does not require RSTP at the gateway VPLS PEs while still using pseudowires to interconnect the VPLS instances located in the two domains. It is supported in both VPLS and PBB-VPLS on the B-VPLS side.

MC-EP expands the single chassis endpoint based on active-standby pseudowires for VPLS, shown in Figure 67.

Figure 67:  HVPLS Resiliency Based on AS Pseudowires 

The active-standby pseudowire solution is appropriate for the scenario when only one VPLS PE (MTUs) needs to be dual-homed to two core PEs (PE1 and PE2). When multiple VPLS domains need to be interconnected, the above solution provides a single point of failure at the MTU-s. The example shown in Figure 68 can be used.

Figure 68:  Multi-Chassis Pseudowire Endpoint for VPLS 

The two gateway pairs, PE3-PE3’ and PE1-PE2, are interconnected using a full mesh of four pseudowires out of which only one pseudowire is active at any time.

The concept of pseudowire endpoint for VPLS provides multi-chassis resiliency controlled by the MC-EP pair, PE3-PE3’ in this example. This scenario, referred to as multi-chassis pseudowire endpoint for VPLS, provides a way to group pseudowires distributed between PE3 and PE3 chassis in a virtual endpoint that can be mapped to a VPLS instance.

The MC-EP inter-chassis protocol is used to ensure configuration and status synchronization of the pseudowires that belong to the same MC-EP group on PE3 and PE3. Based on the information received from the peer shelf and the local configuration, the master shelf decides on which pseudowire will become active.

The MC-EP solution is built around the following components:

3.2.10.3.2. MC-EP Passive Mode

The MC-EP mechanisms are built to minimize the possibility of loops. It is possible that human error could create loops through the VPLS service. One way to prevent loops is to enable the MAC move feature in the gateway PEs (PE3, PE3', PE1, and PE2).

An MC-EP passive mode can also be used on the second PE pair, PE1 and PE2, as a second layer of protection to prevent any loops from occurring if the operator introduces operational errors on the MC-EP PE3, PE3 pair. An example is shown in Figure 69.

Figure 69:  MC-EP in Passive Mode 

When in passive mode, the MC-EP peers stay dormant as long as one active pseudowire is signaled from the remote end. If more than one pseudowire belonging to the passive MC-EP becomes active, the PE1 and PE2 pair applies the MC-EP selection algorithm to select the best choice and blocks all others. No signaling is sent to the remote pair to avoid flip-flop behavior. A trap is generated each time MC-EP in passive mode activates. Every occurrence of this kind of trap should be analyzed by the operator as it is an indication of possible misconfiguration on the remote (active) MC-EP peering.

For the MC-EP passive mode to work, the pseudowire status signaling for active/standby pseudowires should be enabled. This requires the following CLI configurations:

For the remote MC-EP PE3, PE3 pair:

config>service>vpls>endpoint# no suppress-standby-signaling

When MC-EP passive mode is enabled on the PE1 and PE2 pair, the following command is always enabled internally, regardless of the actual configuration:

config>service>vpls>endpoint no ignore-standby-signaling

3.2.10.4. Support for Single Chassis Endpoint Mechanisms

In cases of SC-EP, there is a consistency check to ensure that the configuration of the member pseudowires is the same. For example, mac-pining, mac-limit, and ignore standby signaling must be the same. In the MC-EP case, there is no consistency check between the member endpoints located on different chassis. The operator must carefully verify the configuration of the two endpoints to ensure consistency.

The following rules apply for suppress-standby-signaling and ignore-standby parameters:

This section also describes how the main mechanisms used for single chassis endpoint are adapted for the MC-EP solution.

3.2.10.4.1. MAC Flush Support in MC-EP

In an MC-EP scenario, failure of a pseudowire or gateway PE will determine activation of one of the next best pseudowires in the MC-EP group. This section describes the MAC flush procedures that can be applied to ensure blackhole avoidance.

Figure 70 shows a pair of PE gateways (PE3 and PE3) running MC-EP toward PE1 and PE2 where F1 and F2 are used to indicate the possible direction of the MAC flush, signaled using T-LDP MAC withdraw message. PE1 and PE2 can only use regular VPLS pseudowires and do not have to use an MC-EP or a regular pseudowire endpoint.

Figure 70:  MAC Flush in the MC-EP Solution 

Regular MAC flush behavior will apply for the LDP MAC withdraw sent over the T-LDP sessions associated with the active pseudowire in the MC-EP; for example, PE3 to PE1. That includes any Topology Change Notification (TCN) events or failures associated with SAPs or pseudowires not associated with the MC-EP.

The following MAC flush behaviors apply to changes in the MC-EP pseudowire selection:

1. If the local PW2 becomes active on PE3:
   1. On PE3, the MACs mapped to PW1 are moved to PW2.
   2. A T-LDP flush-all-but-mine message is sent toward PE2 in the F2 direction and is propagated by PE2 in the local VPLS mesh.
   3. No MAC flush is sent in the F1 direction from PE3.
2. If one of the pseudowires on the pair PE3 becomes active; for example, PW4:
   1. On PE3, the MACs mapped to PW1 are flushed, the same as for a regular endpoint.
   2. PE3 must be configured with send-flush-on-failure to send a T-LDP flush-all-from-me message toward VPLS mesh in the F1 direction.
   3. PE3 sends a T-LDP flush-all-but-mine message toward PE2 in the F2 direction, which is propagated by PE2 in the local VPLS mesh. When MC-EP is in passive mode and the first spoke becomes active, a no mac flush-all-but-mine message will be generated.

3.2.10.5. Using B-VPLS for Increased Scalability and Reduced Convergence Times

The PBB-VPLS solution can be used to improve scalability of the solution and to reduce convergence time. If PBB-VPLS is deployed starting at the edge PEs, the gateway PEs will contain only B-VPLS instances. The MC-EP procedures described for regular VPLS apply.

PBB-VPLS can also be enabled just on the gateway MC-EP PEs, as shown in Figure 71.

Figure 71:  MC-EP with B-VPLS 

Multiple I-VPLS instances may be used to represent in the gateway PEs the customer VPLS instances using the PBB-VPLS M:1 model described in the PBB section. A backbone VPLS (B-VPLS) is used in this example to administer the resiliency for all customer VPLS instances at the domain borders. Just one MC-EP is required to be configured in the B-VPLS to address hundreds or even thousands of customer VPLS instances. If load balancing is required, multiple B-VPLS instances may be used to ensure even distribution of the customers across all the pseudowires interconnecting the two domains. In this example, four B-VPLSs will be able to load share the customers across all four possible pseudowire paths.

The use of MC-EP with B-VPLS is strictly limited to cases where VPLS mesh exists on both sides of a B-VPLS. For example, active/standby pseudowires resiliency in the I-VPLS context where PE3 and PE3’ are PE-rs cannot be used because there is no way to synchronize the active/standby selection between the two domains.

For a similar reason, MC-LAG resiliency in the I-VPLS context on the gateway PEs participating in the MC-EP (PE3 and PE3’) should not be used.

For the PBB topology in Figure 71, block-on-mesh-failure in the I-VPLS domain will not have any effect on the B-VPLS MC-EP side. That is because mesh failure in one I-VPLS should not affect other I-VPLSs sharing the same B-VPLS.

3.2.10.6. MAC Flush Additions for PBB VPLS

The scenario shown in Figure 72 is used to define the blackholing problem in PBB-VPLS using MC-EP.

Figure 72:  MC-EP with B-VPLS Failure Scenario 

In the topology shown in Figure 72, PEA and PEB are regular VPLS PEs participating in the VPLS mesh deployed in the metro and WAN region, respectively. As the traffic flows between CEs with C-MAC X and C-MAC Y, the FDB entries in PEA, PE3, PE1 and PEB are installed. An LDP flush-all-but-mine message will be sent from PE3 to PE2 to clear the B-VPLS FDBs. The traffic between C-MAC X and C-MAC Y will be blackholed as long as the entries from the VPLS and I-VPLS FDBs along the path are not removed. This may take as long as 300 seconds, the usual aging timer used for MAC entries in a VPLS FDB.

A MAC flush is required in the I-VPLS space from PBB PEs to PEA and PEB to avoid blackholing in the regular VPLS space.

In the case of a regular VPLS, the following procedure is used:

For consistency, a similar procedure is used for the B-VPLS case as shown in Figure 73.

Figure 73:  MC-EP with B-VPLS MAC Flush Solution 

In this example, the MC-EP activates B-VPLS PW4 because of either a link/node failure or because of an MC-EP selection re-run that affected the previously active PW1. As a result, the endpoint on PE3 containing PW1 goes down.

The following steps apply:

Other failure scenarios are addressed using the same or a subset of the above steps:

For an SC/MC endpoint configured in a B-VPLS, failure/deactivation of the active pseudowire member always generates a local MAC flush of all the B-MAC associated with the pseudowire. It never generates a MAC move to the newly active pseudowire even if the endpoint stays up. That is because in SC-EP/MC-EP topology, the remote PE might be the terminating PBB PE and may not be able to reach the B-MAC of the other remote PE. Therefore, connectivity between them exists only over the regular VPLS mesh.

For the same reasons, Nokia recommends that static B-MAC not be used on SC/MC endpoints.

3.2.11.1. STP-based Redundant Access to VPLS

Figure 74:  Dual-homed MTUs in Two-Tier Hierarchy H-VPLS  

In the configuration shown in Figure 74, STP is activated on the MTU and two PEs in order to resolve a potential loop. STP only needs to run in a single VPLS instance, and the results of the STP calculations are applied to all VPLSs on the link.

In this configuration, the scope of the STP domain is limited to MTU and PEs, while any topology change needs to be propagated in the whole VPLS domain including mesh SDPs. This is done by using so-called MAC-flush messages defined by RFC 4762. In the case of STP as an loop resolution mechanism, every TCN received in the context of an STP instance is translated into an LDP-MAC address withdrawal message (also referred to as a MAC-flush message) requesting to clear all FDB entries except the ones learned from the originating PE. Such messages are sent to all PE peers connected through SDPs (mesh and spoke) in the context of VPLS services, which are managed by the specified STP instance.

3.2.11.2. Redundant Access to VPLS Without STP

The Nokia implementation also includes alternative methods for providing a redundant access to Layer 2 services, such as MC-LAG, MC-APS, or MC-Ring. Also in this case, the topology change event needs to be propagated into the VPLS topology in order to provide fast convergence. The topology change propagation and its corresponding MAC flush processing in a VPLS service without STP is described in Dual Homing to a VPLS Service.

3.2.12.1. VPLS Applicability — Block on VPLS a Failure

This feature is used in VPLS to enhance the existing BGP MH solution by providing a block-on-group failure function similar to the block-on-mesh failure feature implemented for LDP VPLS. On the PE selected as the Designated Forwarder (DF), if the rest of the VPLS endpoints fail (pseudowire spokes/pseudowire mesh and/or SAPs), there is no path forward for the frames sent to the MH site selected as DF. The status of the VPLS endpoints, other than the MH site, is reflected by bringing down/up the objects associated with the MH site.

Support for the feature is provided initially in VPLS and B-VPLS instance types for LDP VPLS, with or without BGP-AD and for BGP VPLS. The following objects may be placed as components of an operational group: BGP VPLS pseudowires, SAPs, spoke-pseudowire, BGP-AD pseudowires. The following objects are supported as monitoring objects: BGP MH site, individual SAP, spoke-pseudowire.

The following rules apply:

There are two steps involved in enabling the block-on-mesh failure feature in a VPLS scenario:

The status of the operational group (oper-group) is dictated by the status of one or more members according to the following rule:

A simple configuration example is described for the case of a BGP VPLS mesh used to interconnect different customer locations. If we assume a customer edge (CE) device is dual-homed to two PEs using BGP MH, the following configuration steps apply:

A simple configuration example follows:

3.2.13.1. Dual Homing to a VPLS Service

Figure 75:  Dual-homed CE Connection to VPLS 

Figure 75 shows a dual-homed connection to VPLS service (PE-A, PE-B, PE-C, PE-D) and operation in case of link failure (between PE-C and L2-B). Upon detection of a link failure, PE-C will send MAC-address-withdraw messages, which will indicate to all LDP peers that they should flush all MAC addresses learned from PE-C. This will lead to a broadcasting of packets addressing affected hosts and relearning in case an alternative route exists.

The message described here is different than the message described in RFC 4762, Virtual Private LAN Services Using LDP Signaling. The difference is in the interpretation and action performed in the receiving PE. According to the standard definition, upon receipt of a MAC withdraw message, all MAC addresses, except the ones learned from the source PE, are flushed. This section specifies that all MAC addresses learned from the source are flushed. This message has been implemented as an LDP address withdraw message with vendor-specific type, length, and value (TLV), and is called the flush-all-from-me message.

The RFC 4762 compliant message is used in VPLS services for recovering from failures in STP (Spanning Tree Protocol) topologies. The mechanism described in this section represents an alternative solution.

The advantage of this approach (as compared to STP-based methods) is that only the affected MAC addresses are flushed and not the full forwarding database. While this method does not provide a mechanism to secure alternative loop-free topology, the convergence time is dependent on the speed that the specified CE device will open an alternative link (L2-B switch in Figure 75) as well as on the speed that PE routers will flush their FDB.

In addition, this mechanism is effective only if PE and CE are directly connected (no hub or bridge) as the mechanism reacts to the physical failure of the link.

3.2.13.2. MC-Ring and VPLS

The use of multi-chassis ring control in a combination with the plain VPLS SAP is supported by the FDB in individual ring nodes, in case the link (or ring node) failure cannot be cleared on the 7750 SR or 7950 XRS.

This combination is not easily blocked in the CLI. If configured, the combination may be functional but the switchover times will be proportional to MAC aging in individual ring nodes and/or to the relearning rate due to downstream traffic.

Redundant plain VPLS access in ring configurations, therefore, exclude corresponding SAPs from the multi-chassis ring operation. Configurations such as M-VPLS can be applied.

3.2.16.1. BGP AD Overview

The BGP protocol establishes neighbor relationships between configured peers. An open message is sent after the completion of the three-way TCP handshake. This open message contains information about the BGP peer sending the message. This message contains Autonomous System Number (ASN), BGP version, timer information, and operational parameters, including capabilities. The capabilities of a peer are exchanged using two numerical values: the Address Family Identifier (AFI) and Subsequent Address Family Identifier (SAFI). These numbers are allocated by the Internet Assigned Numbers Authority (IANA). BGP AD uses AFI 65 (L2VPN) and SAFI 25 (BGP VPLS). The complete list of allocations are at: http://www.iana.org/assignments/address-family-numbers and SAFI http://www.iana.org/assignments/safi-namespace.

3.2.16.2. Information Model

Following the establishment of the peer relationship, the discovery process begins as soon as a new VPLS service instance is provisioned on the PE.

Two VPLS identifiers are used to indicate the VPLS membership and the individual VPLS instance:

To advertise this information, BGP AD employs a simplified version of the BGP VPLS NLRI where just the RD and the next four bytes are used to identify the VPLS instance. There is no need for Label Block and Label Size fields as T-LDP will signal the service labels later on.

The format of the BGP AD NLRI is very similar to the one used for IP VPN, as shown in Figure 78. The system IP may be used for the last four bytes of the VSI ID, further simplifying the addressing and the provisioning process.

Figure 78:  BGP AD NLRI versus IP VPN NLRI 

Network Layer Reachability Information (NLRI) is exchanged between BGP peers indicating how to reach prefixes. The NLRI is used in the Layer 2 VPN case to tell PE peers how to reach the VSI, rather than specific prefixes. The advertisement includes the BGP next hop and a route target (RT). The BGP next hop indicates the VSI location and is used in the next step to determine which signaling session is used for pseudowire signaling. The RT, also coded as an extended community, can be used to build a VPLS full mesh or an HVPLS hierarchy through the use of BGP import/export policies.

BGP is only used to discover VPN endpoints and the corresponding far-end PEs. It is not used to signal the pseudowire labels. This task remains the responsibility of targeted-LDP (T-LDP).

3.2.16.3. FEC Element for T-LDP Signaling

Two LDP FEC elements are defined in RFC 4447, PW Setup & Maintenance Using LDP. The original pseudowire-ID FEC element 128 (0x80) employs a 32-bit field to identify the virtual circuit ID and was used extensively in the initial VPWS and VPLS deployments. The simple format is easy to understand but does not provide the required information model for BGP auto-discovery function. To support BGP AD and other new applications, a new Layer 2 FEC element, the generalized FEC (0x81), is required.

The generalized pseudowire-ID FEC element has been designed for auto-discovery applications. It provides a field, the address group identifier (AGI), that is used to signal the membership information from the VPLS-ID. Separate address fields are provided for the source and target address associated with the VPLS endpoints, called the Source Attachment Individual Identifier (SAII) and Target Attachment Individual Identifier (TAII), respectively. These fields carry the VSI ID values for the two instances that are to be connected through the signaled pseudowire.

The detailed format for FEC 129 is shown in Figure 79.

Figure 79:  Generalized Pseudowire-ID FEC Element 

Each of the FEC fields are designed as a sub-TLV equipped with its own type and length, providing support for new applications. To accommodate the BGP AD information model, the following FEC formats are used:

3.2.16.4. BGP-AD and Target LDP (T-LDP) Interaction

BGP is responsible for discovering the location of VSIs that share the same VPLS membership. LDP protocol is responsible for setting up the pseudowire infrastructure between the related VSIs by exchanging service-specific labels between them.

Once the local VPLS information is provisioned in the local PE, the related PEs participating in the same VPLS are identified through BGP AD exchanges. A list of far-end PEs is generated and will trigger the creation, if required, of the necessary T-LDP sessions to these PEs and the exchange of the service-specific VPN labels. The steps for the BGP AD discovery process and LDP session establishment and label exchange are shown in Figure 80.

Figure 80:  BGP-AD and T-LDP Interaction 

Key:

3.2.16.5. SDP Usage

Service Access Points (SAPs) are linked to transport tunnels using Service Distribution Points (SDPs). The service architecture allows services to be abstracted from the transport network.

MPLS transport tunnels are signaled using the Resource Reservation Protocol (RSVP-TE) or by the Label Distribution Protocol (LDP). The capability to automatically create an SDP only exists for LDP-based transport tunnels. Using a manually provisioned SDP is available for both RSVP-TE and LDP transport tunnels. Refer to the appropriate 7450 ESS, 7750 SR, 7950 XRS, and VSR MPLS Guide for more information about MPLS, LDP, and RSVP.

GRE transport tunnels use GRE encapsulation and can be used with manually provisioned or auto created SDPs.

3.2.16.6. Automatic Creation of SDPs

When BGP AD is used for LDP VPLS, with an LDP or GRE transport tunnel, there is no requirement to manually create an SDP. The LDP or GRE SDP can be automatically instantiated using the information advertised by BGP AD. This simplifies the configuration on the service node.

The use of an automatically created GRE tunnel is enabled by creating the PW template used within the service with the parameter auto-gre-sdp. The GRE SDP and SDP binding is created after a matching BGP route has been received.

Enabling LDP on the IP interfaces connecting all nodes between the ingress and the egress builds transport tunnels based on the best IGP path. LDP bindings are automatically built and stored in the hardware. These entries contain an MPLS label pointing to the best next hop along the best path toward the destination.

When two endpoints need to connect and no SDP exists, a new SDP will automatically be constructed. New services added between two endpoints that already have an automatically created SDP will be immediately used; no new SDP will be constructed. The far-end information is learned from the BGP next hop information in the NLRI. When services are withdrawn with a BGP_Unreach-NLRI, the automatically established SDP will remain up while at least one service is connected between those endpoints. An automatically created SDP will be removed and the resources released when the only or last service is removed.

The service provider has the option of associating the auto-discovered SDP with a split horizon group using the pw-template-binding option, to control the forwarding between pseudowires and to prevent Layer 2 service loops.

An auto-discovered SDP using a pw-template-binding without a split horizon group configured will have similar traffic flooding behavior as a spoke-SDP.

3.2.16.7. Manually Provisioned SDP

The carrier is required to manually provision the SDP if they create transport tunnels using RSVP-TE. Operators have the option to choose a manually configured SDP if they use LDP as the tunnel signaling protocol. The functionality is the same regardless of the signaling protocol.

Creating a BGP AD-enabled VPLS service on an ingress node with the manually provisioned SDP option causes the tunnel manager to search for an existing SDP that connects to the far-end PE. The far-end IP information is learned from the BGP next hop information in the NLRI. If a single SDP exists to that PE, it is used. If no SDP is established between the two endpoints, the service will remain down until a manually configured SDP becomes active.

When multiple SDPs exist between two endpoints, the tunnel manager will select the appropriate SDP. The algorithm will prefer SDPs with the best (lower) metric. If there are multiple SDPs with equal metrics, the operational state of the SDPs with the best metric will be considered. If the operational state is the same, the SDP with the higher SDP-ID will be used. If an SDP with a preferred metric is found with an operational state that is not active, the tunnel manager will flag it as ineligible and restart the algorithm.

3.2.16.8. Automatic Instantiation of Pseudowires (SDP Bindings)

The choice of manual or auto-provisioned SDPs has limited impact on the amount of required provisioning. Most of the savings are achieved through the automatic instantiation of the pseudowire infrastructure (SDP bindings). This is achieved for every auto-discovered VSI through the use of the pseudowire template concept. Each VPLS service that uses BGP AD contains the pw-template-binding option defining specific Layer 2 VPN parameters. This command references a PW template, which defines the pseudowire parameters. The same PW template may be referenced by multiple VPLS services. As a result, changes to these pseudowire templates have to be treated with caution as they may impact many customers simultaneously.

The Nokia implementation provides for safe handling of pseudowire templates. Changes to the pseudowire templates are not automatically propagated. Tools are provided to evaluate and distribute the changes. The following command is used to distribute changes to a PW template at the service level to one or all services that use that template:

PERs-4# tools perform service id 300 eval-pw-template 1 allow-service-impact

If the service ID is omitted, all services will be updated. The type of change made to the PW template will influence how the service is impacted:

Both of these changes are service affecting. Other changes will not be service affecting.

3.2.16.9. Mixing Statically Configured and Auto-Discovered Pseudowires in a VPLS

The services implementation allows for manually provisioned and auto-discovered pseudowire (SDP bindings) to coexist in the same VPLS instance (for example, both FEC128 and FEC 129 are supported). This allows for gradual introduction of auto-discovery into an existing VPLS deployment.

As FEC 128 and 129 represent different addressing schemes, it is important to ensure that only one is used at any time between the same two VPLS instances. Otherwise, both pseudowires may become active causing a loop that might adversely impact the correct functioning of the service. It is recommended that FEC128 pseudowire be disabled as soon as the FEC129 addressing scheme is introduced in a portion of the network. Alternatively, RSTP may be used during the migration as a safety mechanism to provide additional protection against operational errors.

3.2.16.10. Resiliency Schemes

The use of BGP AD on the network side, or in the backbone, does not affect the different resiliency schemes Nokia has developed in the access network. This means that both Multi-Chassis Link Aggregation (MC-LAG) and Management-VPLS (M-VPLS) can still be used.

BGP AD may coexist with Hierarchical-VPLS (H-VPLS) resiliency schemes (for example, dual-homed MTUs devices to different PE-rs nodes) using existing methods (M-VPLS and statically configured active/standby pseudowire endpoint).

If provisioned SDPs are used by BGP AD, M-VPLS may be employed to provide loop avoidance. However, it is currently not possible to auto-discover active/standby pseudowires and to instantiate the related endpoint.

3.2.17.1. Pseudowire Signaling Details

The pseudowire is set up using the following NLRI fields:

This BGP update is telling the other PEs that accept the RT: reach me (VE-ID = x), use a pseudowire label of LB + VE-ID – VBO using the BGP NLRI for which VBO =< local VE-ID < VBO + VBS.

Following is an example of how this algorithm works, assuming PE1 has VE-ID 7 configured:

Assuming that VE-ID = 10 is configured in another PE4, the following procedure applies:

In addition to the pseudowire label information, the Layer2 Info Extended Community attribute must be included in the BGP update for BGP VPLS to signal the attributes of all the pseudowires that converge toward the originator VPLS PE.

The format is as follows:

The meaning of the fields are as follows:

Figure 82 shows the detailed format for the control flags bit vector.

Figure 82:  Control Flag Bit Vector Format 

The following bits in the control flags are defined as follows:

Following are the events that set the D-bit to 1 to indicate VSI down status in the BGP update message sent out from a PE:

The following events do not set the D-bit to 1:

3.2.17.2. Supported VPLS Features

BGP VPLS includes support for a new type of pseudowire signaling based on MP-BGP, based on the existing VPLS instance; therefore, it inherits all the existing Ethernet switching functions.

The use of an automatically created GRE tunnel is enabled by creating the PW template used within the service with the parameter auto-gre-sdp. The GRE SDP and SDP binding is created after a matching BGP route has been received.

Following are some of the most important VPLS features ported to BGP VPLS:

3.2.19.1. Information Model and Required Extensions to L2VPN NLRI

VPLS Multi-homing using BGP-MP expands on the BGP AD and BGP VPLS provisioning model. The addressing for the multi-homed site is still independent from the addressing for the base VSI (VSI-ID or, respectively, VE-ID). Every multi-homed CE is represented in the VPLS context through a site-ID, which is the same on the local PEs. The site-ID is unique within the scope of a VPLS. It serves to differentiate between the multi-homed CEs connected to the same VPLS Instance (VSI). For example, in Figure 84, CE5 will be assigned the same site-ID on both PE1 and PE4. For the same VPLS instance, different site-IDs are assigned for multi-homed CE5 and CE6; for example, site-ID 5 is assigned for CE5 and site-ID 6 is assigned for CE6. The single-homed CEs (CE1, 2, 3, and 4) do not require allocation of a multi-homed site-ID. They are associated with the addressing for the base VSI, either VSI-ID or VE-ID.

The new information model required changes to the BGP usage of the NLRI for VPLS. The extended MH NLRI for Multi-Homed VPLS is compared with the BGP AD and BGP VPLS NLRIs in Figure 84.

Figure 84:  BGP MH-NLRI for VPLS Multi-Homing 

The BGP VPLS NLRI described in RFC 4761 is used to carry a 2-byte site-ID that identifies the MH site. The last seven bytes of the BGP VPLS NLRI used to instantiate the pseudowire are not used for BGP-MH and are zeroed out. This NLRI format translates into the following processing path in the receiving VPLS PE:

The processing procedures described in this section start from the above identification of the BGP update as not destined for pseudowire signaling.

The RD ensures that the NLRIs associated with a certain site-ID on different PEs are seen as different by any of the intermediate BGP nodes (RRs) on the path between the multi-homed PEs. That is, different RDs must be used on the MH PEs every time an RR or an ASBR is involved to guarantee the MH NLRIs reach the PEs involved in VPLS MH.

The L2-Info extended community from RFC 4761 is used in the BGP update for MH NLRI to initiate a MAC flush for blackhole avoidance, to indicate the operational and admin status for the MH site or the DF election status.

After the pseudowire infrastructure between VSIs is built using either RFC 4762, Virtual Private LAN Service (VPLS) Using Label Distribution Protocol (LDP) Signaling, or RFC 4761 procedures, or a mix of pseudowire signaling procedures, on activation of a multi-homed site, an election algorithm must be run on the local and remote PEs to determine which site will be the designated forwarder (DF). The end result is that all the related MH sites in a VPLS will be placed in standby except for the site selected as DF. Nokia BGP-based multi-homing solution uses the DF election procedure described in the IETF working group document draft-ietf-bess-vpls-multihoming-01. The implementation allows the use of BGP local preference and the received VPLS preference but does not support setting the VPLS preference to a non-zero value.

3.2.19.2. Supported Services and Multi-Homing Objects

This feature is supported for the following services:

The following access objects can be associated with MH Site:

3.2.19.3. Blackhole Avoidance

Blackholing refers to the forwarding of frames to a PE that is no longer carrying the designated forwarder. This could happen for traffic from:

Changes in DF election results or MH site status must be detected by all of the above network elements to provide for Blackhole Avoidance.

3.2.19.4. BGP Multi-Homing for VPLS Inter-Domain Resiliency

BGP MH for VPLS can be used to provide resiliency between different VPLS domains. An example of a multi-homing topology is shown in Figure 85.

Figure 85:  BGP MH Used in an HVPLS Topology 

LDP VPLS domains are interconnected using a core VPLS domain, either BGP VPLS or LDP VPLS. The gateway PEs, for example PE2 and PE3, are running BGP multi-homing where one MH site is assigned to each of the pseudowires connecting the access PE, PE7, and PE8 in this example.

Alternatively, the MH site can be associated with multiple access pseudowires using an access SHG. The config>service>vpls>site>failed-threshold command can be used to indicate the number of pseudowire failures that are required for the MH site to be declared down.

3.2.20.1. IGMP Snooping for VPLS

When IGMP snooping is enabled in a VPLS service, IGMP messages received on SAPs and SDPs are snooped in order to determine the scope of the flooding for a specified stream or (S,G). IGMP snooping operates in a proxy mode, where the system summarizes upstream IGMP reports and responds to downstream queries. See “IGMP Snooping” in the 7450 ESS, 7750 SR, and VSR Triple Play Service Delivery Architecture Guide for a description of IGMP snooping.

Streams are sent to all SAPs and SDPs on which there is a multicast router (either discovered dynamically from received query messages or configured statically using the mrouter-port command) and on which an active join for that stream has been received. The mrouter port configuration adds a (*,*) entry into the MFIB, which causes all groups (and IGMP messages) to be sent out of the respective object and causes IGMP messages received on that object to be discarded.

Directly-connected multicast sources are supported when IGMP snooping is enabled.

IGMP snooping is enabled at the service level.

IGMP is not supported in the following:

3.2.20.2. MLD Snooping for VPLS

MLD snooping is an IPv6 version of IGMP snooping. The guidelines and procedures are similar to IGMP snooping as previously described. However, MLD snooping uses MAC-based forwarding. See MAC-Based IPv6 Multicast Forwarding for more information. Directly connected multicast sources are supported when MLD snooping is enabled.

MLD snooping is enabled at the service level and is not supported in the following services:

MLD snooping is not supported under the following forms of default SAP:

MLD snooping is not supported in a VPLS service configured with a connection profile VLAN SAP.

3.2.20.3. PIM Snooping for VPLS

PIM snooping for VPLS allows a VPLS PE router to build multicast states by snooping PIM protocol packets that are sent over the VPLS. The VPLS PE then forwards multicast traffic based on the multicast states. When all receivers in a VPLS are IP multicast routers running PIM, multicast forwarding in the VPLS is efficient when PIM snooping for VPLS is enabled.

Because of PIM join/prune suppression, in order to make PIM snooping operate over VPLS pseudowires, two options are available: plain PIM snooping and PIM proxy. PIM proxy is the default behavior when PIM snooping is enabled for a VPLS.

PIM snooping is supported for both IPv4 and IPv6 multicast by default and can be configured to use SG-based forwarding (see IPv6 Multicast Forwarding for more information).

Directly connected multicast sources are supported when PIM snooping is enabled.

The following restrictions apply to PIM snooping:

3.2.20.4. IPv6 Multicast Forwarding

When MLD snooping or PIM snooping for IPv6 is enabled, the forwarding of IPv6 multicast traffic is MAC-based; see MAC-Based IPv6 Multicast Forwarding for more information.

The operation with PIM snooping for IPv6 can be changed to SG-based forwarding; see SG-Based IPv6 Multicast Forwarding for more information.

The following command configures the IPv6 multicast forwarding mode with the default being mac-based:

The forwarding mode can only be changed when PIM snooping for IPv6 is disabled.

3.2.20.5. PIM and IGMP/MLD Snooping Interaction

When both PIM snooping for IPv4 and IGMP snooping are enabled in the same VPLS service, multicast traffic is forwarded based on the combined multicast forwarding table. When PIM snooping is enabled, IGMP queries are forwarded but not snooped, consequently the IGMP querier needs to be seen either as a PIM neighbor in the VPLS service or the SAP towards it configured as an IGMP mrouter port.

There is no interaction between PIM snooping for IPv6 and PIM snooping for IPv4/IGMP snooping when all are enabled within the same VPLS service. The configurations of PIM snooping for IPv6 and MLD snooping are mutually exclusive.

When PIM snooping is enabled within a VPLS service, all IP multicast traffic and flooded PIM messages (these include all PIM snooped messages when not in PIM proxy mode and PIM hellos when in PIM proxy mode) will be sent to any SAP or SDP binding configured with an IGMP-snooping mrouter port. This will occur even without IGMP-snooping enabled but is not supported in a BGP-VPLS or M-VPLS service.

3.2.20.6. Multi-Chassis Synchronization for Layer 2 Snooping States

To achieve a faster failover in scenarios with redundant active/standby routers performing Layer 2 multicast snooping, it is possible to synchronize the snooping state from the active router to the standby router, so that if a failure occurs the standby router has the Layer 2 multicast snooped states and is able to forward the multicast traffic immediately. Without this capability, there would be a longer delay in re-establishing the multicast traffic path due to having to wait for the Layer 2 states to be snooped.

Multi-chassis synchronization (MCS) is enabled per peer router and uses a sync-tag, which is configured on the objects requiring synchronization on both of the routers. This allows MCS to map the state of a set of objects on one router to a set of objects on the other router. Specifically, objects relating to a sync-tag on one router are backed up by, or are backing up, the objects using the same sync-tag on the other router (the state is synchronized from the active object on one router to its backup objects on the standby router).

The object type must be the same on both routers; otherwise, a mismatch error is reported. The same sync-tag value can be reused for multiple peer/object combinations, where each combination represents a different set of synchronized objects; however, a sync-tag cannot be configured on the same object to more than one peer.

The sync-tag is configured per port and can relate to a specific set of dot1q or QinQ VLANs on that port, as follows.

For IGMP snooping and PIM snooping for IPv4 to work correctly with MCS on QinQ ports using x.* SAPs, one of the following must be true:

3.2.20.7. VPLS Multicast-Aware High Availability Features

The following features are High Availability capable:

3.3.1.1. Assigning a Service Name to a VPLS Service

When a service name is applied to any service context, the name and service ID association is registered with the system. A service name cannot be assigned to more than one service ID.

Special consideration is given to a service name that is assigned to a VPLS service that has the config>service>vpls>allow-ip-int-bind command enabled. If a name is applied to the VPLS service while the flag is set, the system will scan the existing IES and VPRN services for an IP interface that is bound to the specified service name. If an IP interface is found, the IP interface will be attached to the VPLS service associated with the name. Only one interface can be bound to the specified name.

If the allow-ip-int-bind command is not enabled on the VPLS service, the system will not attempt to resolve the VPLS service name to an IP interface. As soon as the allow-ip-int-bind flag is configured on the VPLS, the corresponding IP interface will be bound and become operational up. There is no need to toggle the shutdown/no shutdown command.

If an IP interface is not currently bound to the service name used by the VPLS service, no action is taken at the time of the service name assignment.

3.3.1.2. Service Binding Requirements

If the defined service ID is created on the system, the system will check to ensure that the service type is VPLS. If the service type is not VPLS or I-VPLS, service creation will not be allowed and the service ID will remain undefined within the system.

If the created service type is VPLS, the IP interface will be eligible to enter the operationally up state.

3.3.1.3. Bound Service Name Assignment

If a bound service name is assigned to a service within the system, the system will first check to ensure the service type is VPLS or I-VPLS. Secondly, the system will ensure that the service is not already bound to another IP interface via the service ID. If the service type is not VPLS or I-VPLS or the service is already bound to another IP interface via the service ID, the service name assignment will fail.

If a single VPLS service ID and service name is assigned to two separate IP interfaces, the VPLS service will not be allowed to enter the operationally up state.

3.3.1.4. Binding a Service Name to an IP Interface

An IP interface within an IES or VPRN service context may be bound to a service name at any time. Only one interface can be bound to a service.

When an IP interface is bound to a service name and the IP interface is administratively up, the system will scan for a VPLS service context using the name and take the following actions:

3.3.1.5. Bound Service Deletion or Service Name Removal

If a VPLS service is deleted while bound to an IP interface, the IP interface will enter the Down: Non-existent svc-ID operational state. If the IP interface was bound to the VPLS service name, the IP interface will enter the Down: Non-existent svc-name operational state. No console warning is generated.

If the created service type is VPLS, the IP interface will be eligible to enter the operationally up state.

3.3.1.6. IP Interface Attached VPLS Service Constraints

Once a VPLS service has been bound to an IP interface through its service name, the service name assigned to the service cannot be removed or changed unless the IP interface is first unbound from the VPLS service name.

A VPLS service that is currently attached to an IP interface cannot be deleted from the system unless the IP interface is unbound from the VPLS service name.

The allow-ip-int-bind flag within an IP interface attached VPLS service cannot be reset. The IP interface must first be unbound from the VPLS service name to reset the flag.

3.3.1.7. IP Interface and VPLS Operational State Coordination

When the IP interface is successfully attached to a VPLS service, the operational state of the IP interface will be dependent upon the operational state of the VPLS service.

The VPLS service remains down until at least one virtual port (SAP, spoke-SDP, or mesh SDP) is operational.

3.3.2.1. Unicast IP Routing into a VPLS Service

The VPLS service MTU and the IP interface MTU parameters may be changed at any time.

3.3.3.1. R-VPLS Specific ARP Cache Behavior

In typical routing behavior, the system uses the IP route table to select the egress interface, and then at the egress forwarding engine, an ARP entry is used to forward the packet to the appropriate Ethernet MAC. With R-VPLS, the egress IP interface may be represented by a multiple egress forwarding engine (wherever the VPLS service virtual ports exist).

To optimize routing performance, the ingress forwarding engine processing has been augmented to perform an ingress ARP lookup in order to resolve which VPLS MAC address the IP frame must be routed toward. This MAC address may be currently known or unknown within the VPLS FDB. If the MAC is unknown, the packet is flooded by the ingress forwarding engine to all egress forwarding engines where the VPLS service exists. When the MAC is known on a virtual port, the ingress forwarding engine forwards the packet to the correct egress forwarding engine. Table 15 describes how the ARP cache and MAC FDB entry states interact at ingress and Table 16 describes the corresponding egress behavior.

3.3.4.1. R-VPLS SAPs Only Supported on Standard Ethernet Ports

The allow-ip-int-bind flag is set (routing support enabled) on a VPLS/I-VPLS service. SAPs within the service can be created on standard Ethernet, and CCAG ports. ATM and POS are not supported.

3.3.4.2. LAG Port Membership Constraints

If a LAG has a non-supported port type as a member, a SAP for the routing-enabled VPLS service cannot be created on the LAG. Once one or more routing enabled VPLS SAPs are associated with a LAG, a non-supported Ethernet port type cannot be added to the LAG membership.

3.3.4.3. R-VPLS Feature Restrictions

When the allow-ip-int-bind flag is set on a VPLS service, the following restrictions apply. The flag also cannot be enabled while any of these features are applied to the VPLS service:

3.3.4.4. Routed I-VPLS Feature Restrictions

The following restrictions apply to routed I-VPLS:

3.3.7.1. VPLS SAP Ingress IP Filter Override

When an IP Interface is attached to a VPLS or an I-VPLS service context, the VPLS SAP provisioned IP filter for ingress routed packets may be optionally overridden in order to provide special ingress filtering for routed packets. This allows different filtering for routed packets and non-routed packets. The filter override is defined on the IP interface bound to the VPLS service name. A separate override filter may be specified for IPv4 and IPv6 packet types.

If a filter for a specified packet type (IPv4 or IPv6) is not overridden, the SAP specified filter is applied to the packet (if defined).

3.3.7.2. IP Interface Defined Egress QoS Reclassification

The SAP egress QoS policy defined forwarding class and profile reclassification rules are not applied to egress routed packets. To allow for egress reclassification, a SAP egress QoS policy ID may be optionally defined on the IP interface that will be applied to routed packets that egress the SAPs on the VPLS or I-VPLS service associated with the IP interface. Both unicast directed and MAC unknown flooded traffic apply to this rule. Only the reclassification portion of the QoS policy is applied, which includes IP precedence or DSCP classification rules and any defined IP match criteria and their associated actions.

The policers and queues defined within the QoS policy applied to the IP interface are not created on the egress SAPs of the VPLS service. Instead, the QoS policy applied to the egress SAPs defines the egress policers and queues that will be used by both routed and non-routed egress packets. The forwarding class mappings defined in the egress SAP’s QoS policy will also define which policer or queue will handle each forwarding class for both routed and non-routed packets.

3.3.7.3. Remarking for VPLS and Routed Packets

The remarking of packets to and from an IP interface in an R-VPLS service corresponds to that supported on IP interface, even though the packets ingress or egress a SAP in the VPLS service bound to the IP service. Specifically, this results in the ability to remark the DSCP/prec for these packets.

Packets ingressing and egressing SAPs in the VPLS service (not routed through the IP interface) support the regular VPLS QoS and, therefore, the DSCP/prec cannot be remarked.

3.3.7.4. IPv4 Multicast Routing

When using IPv4 Multicast routing, the following are not supported:

3.3.7.5. R-VPLS Supported Routing-related Protocols

The following protocols are supported on IP interfaces bound to a VPLS service:

3.3.7.6. Spanning Tree and Split Horizon

A R-VPLS context supports all spanning tree and split horizon capabilities that a non-R-VPLS service supports.

3.4.4.1. Configure the MVRP Infrastructure using an M-VPLS Context

The following provisioning steps apply:

3.4.4.2. Instantiate Related VLAN FDBs and Trunks in MVRP Scope

This requires the configuration in the M-VPLS, under vpls-group, of the following attributes: VLAN ranges, vpls-template and vpls-sap-template bindings. As soon as the VPLS group is enabled, the configured attributes are used to auto-instantiate, on a per-VLAN basis, a VPLS FDB and related SAPs in the switches and on the “trunk ports” specified in the M-VPLS context. The trunk ports are ports associated with an M-VPLS SAP not configured as an end-station.

The following procedure is used:

The above procedure may be used outside of the MVRP context to pre-provision a large number of VPLS contexts that share the same infrastructure and attributes.

The MVRP control of the auto-instantiated services can be enabled using the mvrp-contrl command under vpls-group:

From an MVRP perspective, these SAPs can be either “full MVRP” or “end-station” interfaces.

A full MVRP interface is a full participant in the local M-VPLS scope:

In an MVRP end-station interface, the attributes registered on that interface have local significance:

The following example describes the M-VPLS configuration required to auto-instantiate the VLAN FDBs and related trunks in non-PBB switches:

A similar M-VPLS configuration may be used to auto-instantiate the VLAN FDBs and related trunks in PBB switches. The vpls-group command is replaced by the end-station command under the downward-facing SAPs as in the following example:

3.4.4.3. MVRP Activation of Service Connectivity

As new Ethernet services are activated, UNI SAPs need to be configured and associated with the VLAN IDs (VPLS instances) auto-created using the procedures described in the previous sections. These UNI SAPs may be located in the same VLAN domain or over a PBB backbone. When UNI SAPs are located in different VLAN domains, an intermediate service translation point must be used at the PBB BEB, which maps the local VLAN ID through an I-VPLS SAP to a PBB ISID. This BEB SAP will be playing the role of an end-station from an MVRP perspective for the local VLAN domain.

This section will discuss how MVRP is used to activate service connectivity between a BEB SAP and a UNI SAP located on one of the switches in the local domain. A similar procedure is used in the case of UNI SAPs configured on two switches located in the same access domain. No end-station configuration is required on the PBB BEB if all the UNI SAPs in a service are located in the same VLAN domain.

The service connectivity instantiation through MVRP is shown in Figure 89.

Figure 89:  Service Instantiation with MVRP - QinQ to PBB Example 

In this example, the UNI and service translation SAPs are configured in the data VPLS represented by the gray circles. This instance and associated trunk SAPs were instantiated using the procedures described in the previous sections. The following configuration steps are involved:

As soon as the first UNI SAP becomes active in the data VPLS on the ES, the associated VLAN value is advertised by MVRP throughout the related M-VPLS context. As soon as the second UNI SAP becomes available on a different switch, or in our example on the PBB BEB, the MVRP proceeds to advertise the associated VLAN value throughout the same M-VPLS. The trunks that experience MVRP declaration and registration in both directions will become active, instantiating service connectivity as represented by the big and small yellow circles shown in the figure.

A hold-time parameter (config>service>vpls>mrp>mvrp>hold-time) is provided in the M-VPLS configuration to control when the end-station or last UNI SAP is considered active from an MVRP perspective. The hold-time controls the amount of MVRP advertisements generated on fast transitions of the end-station or UNI SAPs.

If the no hold-time setting is used:

If a non-zero “hold-time” setting is used:

For QinQ end-station SAPs, only no hold-time setting is allowed.

Only the following PBB Epipe and I-VPLS SAP types are eligible to activate MVRP declarations:

Examples of steps required to activate service connectivity for VLAN 100 using MVRP follows.

In the data VPLS instance (VLAN 100) controlled by MVRP, on the QinQ switch:

In I-VPLS on PBB BEB:

3.4.4.4. MVRP Control Plane

MVRP is based on the IEEE 802.1ak MRP specification where STP is the supported method to be used for loop avoidance in a native Ethernet environment. M-VPLS and the associated MSTP (or P-MSTP) control plane provides the loop avoidance component in the Nokia implementation. Nokia MVRP may also be used in a non- MSTP, loop free topology.

3.4.4.5. STP-MVRP Interaction

Table 17 shows the expected interaction between STP (MSTP or P-MSTP) and MVRP.

Notes:

3.4.5.1. VPLS E-Tree Services Overview

The VPLS E-Tree service offers a VPLS service with Root and Leaf designated access SAPs and SDP bindings, which prevent any traffic flow from leaf to leaf directly. With a VPLS E-Tree, the split horizon group capability is inherent for leaf SAPs (or SDP bindings) and extends to all the remote PEs that are part of the same VPLS E-Tree service. This feature is based on IETF Draft draft-ietf-l2vpn-vpls-pe-etree.

A VPLS E-Tree service may support an arbitrary number of leaf access (leaf-ac) interfaces, root access (root-ac) interfaces, and root-leaf tagged (root-leaf-tag) interfaces. Leaf-ac interfaces are supported on SAPs and SDP binds and can only communicate with root-ac interfaces (also supported on SAPs and SDP binds). Leaf-ac to leaf-ac communication is not allowed. Root-leaf-tag interfaces (supported on SAPs and SDP bindings) are tagged with root and leaf VIDs to allow remote VPLS instances to enforce the E-Tree forwarding.

Figure 90 shows a network with two root-ac interfaces and several leaf-ac SAPs (also could be SDPs). The figure indicates two VIDs in use to each service within the service with no restrictions on the AC interfaces. The service guarantees no leaf-ac to leaf-ac traffic.

Figure 90:  E-Tree Service  

3.4.5.2. Leaf-ac and Root-ac SAPs

Figure 91 shows the terminology used for E-Tree in IETF Draft draft-ietf-l2vpn-vpls-pe-etree and a mapping to SR OS terms.

An Ethernet service access SAP is characterized as either a leaf-ac or a root-ac for a VPLS E-Tree service. As far as SR OS is concerned, these are normal SAPs with either no tag (Null), priority tag, or dot1q or QinQ encapsulation on the frame. Functionally, a root-ac is a normal SAP and does not need to be differentiated from the regular SAPs except that it will be associated with a root behavior in a VPLS E-Tree.

Leaf-ac SAPs have restrictions; for example, a SAP configured for a leaf-ac can never send frames to another leaf-ac directly (local) or through a remote node. Leaf-ac SAPs on the same VPLS instance behave as if they are part of a split horizon group (SHG) locally. Leaf-ac SAPs that are on other nodes need to have the traffic marked as originating “from a Leaf” in the context of the VPLS service when carried on PWs and SAPs with tags (VLANs).

Root-ac SAPs on the same VPLS can talk to any root-ac or leaf-ac.

Figure 91:  Mapping PE Model to VPLS Service 

3.4.5.3. Leaf-ac and Root-ac SDP Binds

Untagged SDP binds for access can also be designated as root-ac or leaf-ac. This type of E-Tree interface is required for devices that do not support E-Tree, such as the 7210 SAS, to enable them to be connected with pseudowires. Such devices are root or leaf only and do not require having a tagged frame with a root or leaf indication.

3.4.5.4. Root-leaf-tag SAPs

Support on root-leaf-tag SAPs requires that the outer VID is overloaded to indicate root and leaf. To support the SR service model for a SAP, the ability to send and receive two different tags on a single SAP has been added. Figure 92 shows the behavior when a root-ac and leaf-ac exchange traffic over a root-leaf-tag SAP. Although the figure shows two SAPs connecting VPLS instances 1 and 2, the CLI will show a single SAP with the format:

sap 2/1/1:25 root-leaf-tag leaf-tag 26 create

Figure 92:  Leaf and Root Tagging dot1q 

The root-leaf-tag SAP performs all of the operations for egress and ingress traffic for both tags (root and leaf):

3.4.5.5. Root-leaf-tag SDP Binds

Typically, in a VPLS environment over MPLS, mesh and spoke-SDP binds interconnect the local VPLS instances to remote PEs. To support VPLS E-Tree, the root and leaf traffic is sent over the SDP bind using a fixed VLAN tag value. The SR OS implementation uses a fixed VLAN ID 1 for root and fixed VLAN ID 2 for leaf. The root and leaf tags are considered a global value and signaling is not supported. The vc-type on root-leaf-tag SDP binds must be VLAN. The vlan-vc-tag command will be blocked in root-leaf-tag SDP-binds.

Figure 93 shows the behavior when leaf-ac or root-ac interfaces exchange traffic over a root-leaf-tag SDP-binding.

Figure 93:  Leaf and Root Tagging PW 

3.4.5.6. Interaction between VPLS E-Tree Services and Other Features

As a general rule, any CPM-generated traffic is always root traffic (STP, OAM, and so on) and any received control plane frame is marked with a root/leaf indication based on which E-Tree interface it arrived at. Some other particular feature interactions are as follows: