One of the most popular of the MPLS applications is called MPLS virtual private networks (VPNs). MPLS VPNs allow a service provider, or even a large enterprise, to offer Layer 3 VPN services. In particular, SPs oftentimes replace older Layer 2 WAN services such as Frame Relay and ATM with an MPLS VPN service. MPLS VPN services enable the possibility for the SP to provide a wide variety of additional services to its customers because MPLS VPNs are aware of the Layer 3 addresses at the customer locations. Additionally, MPLS VPNS can still provide the privacy inherent in Layer 2 WAN services.
MPLS VPNs use MPLS unicast IP forwarding inside the SP's network, with additional MPLS-aware features at the edge between the provider and the customer. Additionally, MPLS VPNs use MP-BGP to overcome some of the challenges when connecting an IP network to a large number of customer IP internetworks—problems that include the issue of dealing with duplicate IP address spaces with many customers.
This section begins by examining some of the problems with providing Layer 3 services and then shows the core features of MPLS that solve those problems.
The Problem: Duplicate Customer Address Ranges
When an SP connects to a wide variety of customers using a Layer 2 WAN service such as Frame Relay or ATM, the SP does not care about the IP addressing and subnets used by those customers. However, in order to migrate those same customers to a Layer 3 WAN service, the SP must learn address ranges from the various customers and then advertise those routes into the SP's network. However, even if the SP wanted to know about all subnets from all its customers, many enterprises use the same address ranges—namely, the private IP network numbers, including the ever-popular network 10.0.0.0.
If you tried to support multiple customers using MPLS unicast IP routing alone, the routers would be confused by the overlapping prefixes, as shown in Figure 19-11. In this case, the network shows five of the SP's routers inside a cloud. Three customers (A, B, and C) are shown, with two customer routers connected to the SP's network. All three customers use network 10.0.0.0, with the three customer sites on the right all using subnet 10.3.3.0/24.
Figure 19-11 The Main Challenge with Supporting Layer 3 VPNs
The first and most basic goal for a Layer 3 VPN service is to allow customer A sites to communicate with customer A sites—and only customer A sites. However, the network in Figure 19-11 fails to meet this goal for several reasons. Because of the overlapping address spaces, several routers would be faced with the dilemma of choosing one customer's route to 10.3.3.0/24 as the best route, and ignoring the route to 10.3.3.0/24 learned from another customer. For example, PE2 would learn about two different 10.3.3.0/24 prefixes. If PE2 chooses one of the two possible routes—for example, if PE2 picked the route to CE-A2 as best—then PE2 could not forward packets to customer B's 10.3.3.0/24 off router CE-B2. Also, a possibly worse effect is that hosts in one customer site may be able to send and receive packets with hosts in another customer's network. Following this same example, hosts in customer B and C sites could forward packets to subnet 10.3.3.0/24, and the routers might forward these packets to customer A's CE-A2 router.
The Solution: MPLS VPNs
The protocols and standards defined by MPLS VPNs solve the problems shown in Figure 19-11 and provide a much larger set of features. In particular, the MPLS VPN RFCs define the concept of using multiple routing tables, called Virtual Routing and Forwarding (VRF) tables, which separate customer routes to avoid the duplicate address range issue. This section defines some key terminology and introduces the basics of MPLS VPN mechanics.
MPLS uses three terms to describe the role of a router when building MPLS VPNs. Note that the names used for the routers in most of the figures in this chapter have followed the convention of identifying the type of router as CE, PE, or P, as listed here.
- Customer edge (CE)—A router that has no knowledge of MPLS protocols and does not send any labeled packets but is directly connected to an LSR (PE) in the MPLS VPN.
- Provider edge (PE)—An LSR that shares a link with at least one CE router, thereby providing function particular to the edge of the MPLS VPN, including IBGP and VRF tables
- Provider (P)—An LSR that does not have a direct link to a CE router, which allows the router to just forward labeled packets, and allows the LSR to ignore customer VPNs' routes
The key to understanding the general idea of how MPLS VPNs work is to focus on the control plane distinctions between PE routers and P routers. Both P and PE routers run LDP and an IGP to support unicast IP routing—just as was described in the first half of this chapter. However, the IGP advertises routes only for subnets inside the MPLS network, with no customer routes included. As a result, the P and PE routers can together label switch packets from the ingress PE to the egress PE.
PEs have several other duties as well, all geared toward the issue of learning customer routes and keeping track of which routes belong to which customers. PEs exchange routes with the connected CE routers from various customers, using either EBGP, RIP-2, OSPF, or EIGRP, noting which routes are learned from which customers. To keep track of the possibly overlapping prefixes, PE routers do not put the routes in the normal IP routing table—instead, PEs store those routes in separate per-customer routing tables, called VRFs. Then the PEs use IBGP to exchange these customer routes with other PEs—never advertising the routes to the P routers. Figure 19-12 shows the control plane concepts.
Figure 19-12 Overview of the MPLS VPN Control Plane
The MPLS VPN data plane also requires more work and thought by the PE routers. The PE routers do not have any additional work to do, with one small exception, as compared with simple unicast IP routing. The extra work for the PE relates to the fact that the MPLS VPN data plane causes the ingress PE to place two labels on the packet, as follows:
- An outer MPLS header (S-bit = 0), with a label value that causes the packet to be label switched to the egress PE
- An inner MPLS header (S-bit = 1), with a label that identifies the egress VRF on which to base the forwarding decision
Figure 19-13 shows a general conceptual view of the two labels and the forwarding process. The figure shows a subset of Figure 19-12, with parts removed to reduce clutter. In this case, a host in customer A on the left side of the figure sends a packet to host 10.3.3.3, located on the right side of the figure.
Figure 19-13 Overview of the MPLS VPN Data Plane
The figure shows the following steps:
- CE1 forwards an unlabeled packet to PE1.
- PE1, having received the packet in an interface assigned to VRF-A, compares the packet's destination (10.3.3.3) to the VRF-A CEF FIB, which is based on VRF-A's routing table. PE1 adds two labels based on the FIB and forwards the labeled packet.
- P1, acting just the same as with unicast IP routing, processes the received labeled packet using its LFIB, which simply causes a label swap. P1 forwards the packet to PE2.
- PE2's LFIB entry for label 2222 lists a pop action, causing PE2 to remove the outer label. PE2's LFIB entry for label 3333, populated based on the VRF for customer A's VPN, also lists a pop action and the outgoing interface. As a result, PE2 forwards the unlabeled packet to CE2.
The control plane and data plane processes described around Figures 19-12 and 19-13 outline the basics of how MPLS VPNs work. Next, the chapter takes the explanations a little deeper with a closer look at the new data structures and control plane processes that support MPLS VPNs.
The MPLS VPN Control Plane
The MPLS VPN control plane defines protocols and mechanisms to overcome the problems created by overlapping customer IP address spaces, while adding mechanisms to add more functionality to an MPLS VPN, particularly as compared to traditional Layer 2 WAN services. To understand the mechanics, you need a good understanding of BGP, IGPs, and several new concepts created by both MP-BGP RFCs and MPLS RFCs. In particular, this section introduces and explains the concepts behind three new concepts created for MPLS VPNs:
- Route Distinguishers (RDs)
- Route Targets (RTs)
The next several pages of text examine these topics in order. While reading the rest of the MPLS VPN coverage in this chapter, note that the text will keep expanding a single example. The example focuses on how the control plane learns about routes to the duplicate customer subnets 10.3.3.0/24 on the right side of Figure 19-12, puts the routes into the VRFs on PE2, and advertises the routes with RDs over to PE1 and then how RTs then dictate how PE1 adds the routes to its VRFs.
Virtual Routing and Forwarding Tables
To support multiple customers, MPLS VPN standards include the concept of a virtual router. This feature, called a VRF table, can be used to store routes separately for different customer VPNs. The use of separate tables solves part of the problems of preventing one customer's packets from leaking into another customer's network due to overlapping prefixes, while allowing all sites in the same customer VPN to communicate.
A VRF exists inside a single MPLS-aware router. Typically, routers need at least one VRF for each customer attached to that particular router. For example, in Figure 19-12, router PE2 connects to CE routers in customers A and B but not in customer C, so PE2 would not need a VRF for customer C. However, PE1 connects to CE routers for three customers, so PE1 will need three different VRFs.
For more complex designs, a PE might need multiple VRFs to support a single customer. Using Figure 19-12 again as an example, PE1 connects to two CEs of customer A (CE-A1 and CE-A4). If hosts near CE-A1 were allowed to access a centralized shared service (not shown in the figure) and hosts near CE-A4 were not allowed access, then PE1 would need two VRFs for customer A—one with routes for the shared service's subnets and one without those routes.
Each VRF has three main components, as follows:
- An IP routing table (RIB)
- A CEF FIB, populated based on that VRF's RIB
- A separate instance or process of the routing protocol used to exchange routes with the CEs that need to be supported by the VRF
For example, Figure 19-14 shows more detail about router PE2 from Figure 19-12, now with MPLS VPNs implemented. In this case, PE2 will use RIP-2 as the IGP to both customer A (router CE-A2) and customer B (router CE-B2). (The choice of routing protocol used from PE-CE is unimportant to the depth of explanations shown here.)
Figure 19-14 Adding Routes Learned from a CE to VRFs on Router PE2
The figure shows three parallel steps that occur with each of the two customers. Note that step 1 for each customer does not occur at the same instant in time, nor does step 2, nor step 3; the figure lists these steps with the same numbers because the same function occurs at each step. The explanation of the steps is as follows:
- The CE router, which has no knowledge of MPLS at all, advertises a route for 10.3.3.0/24 as normal—in this case with RIP-2.
- In the top instance of step 2, the RIP-2 update arrives on PE3's S0/1/0, which has been assigned to customer A's VRF, VRF-A. PE2 uses a separate RIP process for each VRF, so PE2's VRF-A RIP process interprets the update. Similarly, the VRF-B RIP process analyzes the update received on S0/1/1 from CE-B2.
- In the top instance of step 3, the VRF-A RIP process adds an entry for 10.3.3.0/24 to the RIB for VRF-A. Similarly, the bottom instance of step 3 shows the RIP process for VRF-B adding a route to prefix 10.3.3.0/24 to the VRF-B RIB.
MP-BGP and Route Distinguishers
Now that PE2 has learned routes from both CE-A2 and CE-B2, PE2 needs to advertise those routes to the other PEs, in order for the other PEs to know how to forward packets to the newly learned subnets. MPLS VPN protocols define the use of IBGP to advertise the routes—all the routes, from all the different VRFs. However, the original BGP specifications did not provide a way to deal with the fact that different customers may use overlapping prefixes.
MPLS deals with the overlapping prefix problem by adding another number in front of the original BGP NLRI (prefix). Each different number can represent a different customer, making the NLRI values unique. To do this, MPLS took advantage of a BGP RFC, called MP-BGP (RFC 4760), which allows for the re-definition of the NLRI field in BGP Updates. This re-definition allows for an additional variable-length number, called an address family, to be added in front of the prefix. MPLS RFC 4364, "BGP/MPLS IP Virtual Private Networks (VPNs)," defines a specific new address family to support IPv4 MPLS VPNs—namely, an MP-BGP address family called Route Distinguishers (RDs).
RDs allow BGP to advertise and distinguish between duplicate IPv4 prefixes. The concept is simple: advertise each NLRI (prefix) as the traditional IPv4 prefix, but add another number (the RD) that uniquely identifies the route. In particular, the new NLRI format, called VPN-V4, has the following two parts:
- A 64-bit RD
- A 32-bit IPv4 prefix
For example, Figure 19-15 continues the story from Figure 19-14, with router PE2 using MP-BGP to advertise its two routes for IPv4 prefix 10.3.3.0/24 to PE1—one from VRF-A and one from VRF-B. The BGP Update shows the new VPN-V4 address family format for the NLRI information, using RD 1:111 to represent VPN-A, and 2:222 to represent VPN-B.
Figure 19-15 Making Prefixes Unique Using an RD
Without the RD as part of the VPN-V4 NLRI, PE1 would have learned about two identical BGP prefixes (10.3.3.0/24) and would have had to choose one of the two as the best route—giving PE1 reachability to only one of the two customer 10.3.3.0/24 subnets. With VPN-V4 NLRI, IBGP advertises two unique NLRI—a 1:111:10.3.3.0 (from VRF-A) and 2:222:10.3.3.0 (from VRF-B). As a result, PE1 keeps both NLRI in its BGP table. The specific steps shown in the figure are explained as follows:
- PE2 redistributes from each of the respective per-VRF routing protocol instances (RIP-2 in this case) into BGP.
- The redistribution process pulls the RD from each respective VRF and includes that RD with all routes redistributed from the VRF's routing table.
- PE3 uses IBGP to advertise these routes to PE1, causing PE1 to know both routes for 10.3.3.0/ 24, each with the differing RD values.
The RD itself is 8 bytes with some required formatting conventions. The first 2 bytes identify which of the three formats is followed. Incidentally, because IOS can tell which of the three formats is used based on the value, the IOS rd VRF subcommand only requires that you type the integer values for the last 6 bytes, with IOS inferring the first 2 bytes (the type) based on the value. The last 6 bytes, as typed in the rd command and seen in show commands, follow one of these formats:
In all three cases, the first value (before the colon) should be either an ASN or an IPv4 address. The second value, after the colon, can be any value you wish. For example, you might choose an RD that lists an LSR's BGP ID using the third format, like 18.104.22.168:100, or you may use the BGP ASN, for example, 432:1.
At this point in the ongoing example, PE1 has learned about the two routes for 10.3.3.0/24—one for VPN-A and one for VPN-B—and the routes are in the BGP table. The next section describes how PE1 then chooses the VRFs into which to add these routes, based on the concept of a Route Target.
One of the most perplexing concepts for engineers, when first learning about MPLS VPNs, is the concept of Route Targets. Understanding the basic question of what RTs do is relatively easy, but understanding why MPLS needs RTs and how to best choose the actual values to use for RTs, can be a topic for long conversation when building an MPLS VPN. In fact, MPLS RTs enable MPLS to support all sorts of complex VPN topologies—for example, allowing some sites to be reachable from multiple VPNs, a concept called overlapping VPNs.
PEs advertise RTs in BGP Updates as BGP Extended Community path attributes (PAs). Generally speaking, BGP extended communities are 8 bytes in length, with the flexibility to be used for a wide variety of purposes. More specifically, MPLS defines the use of the BGP Extended Community PA to encode one or more RT values.
RT values follow the same basic format as the values of an RD. However, note that while a particular prefix can have only one RD, that same prefix can have one or more RTs assigned to it.
To best understand how MPLS uses RTs, first consider a more general definition of the purpose of RTs, followed by an example of the mechanics by which PEs use the RT:
- MPLS uses Route Targets to determine into which VRFs a PE places IBGP-learned routes.
Figure 19-16 shows a continuation of the same example in Figures 19-14 and 19-15, now focusing on how the PEs use the RTs to determine into which VRFs a route is added. In this case, the figure shows an export RT—a configuration setting in VRF configuration mode—with a different value configured for VRF-A and VRF-B, respectively. PE1 shows its import RT for each VRF—again a configuration setting in VRF configuration mode—which allows PE1 to choose which BGP table entries it pulls into each VRF's RIB.
Figure 19-16 The Mechanics of the MPLS Route Target
The figure has a lot of details, but the overall flow of concepts is not terribly difficult. Pay particular attention to the last two steps. Following the steps in the figure:
- The two VRFs on PE2 are configured with an export RT value.
- Redistribution out of the VRF into BGP occurs.
- This step simply notes that the export process—the redistribution out of the VRF into BGP—sets the appropriate RT values in PE2's BGP table.
- PE2 advertises the routes with IBGP.
- PE1 examines the new BGP table entries and compares the RT values to the configured import RT values, which identifies which BGP table entries should go into which VRF.
- PE1 redistributes routes into the respective VRFs, specifically the routes whose RTs match the import RT configured in the VRFs, respectively.
Each VRF needs to export and import at least one RT. The example in Figure 19-16 shows only one direction: exporting on the right (PE2) and importing on the left (PE1). However, PE2 needs to know the routes for the subnets connected to CE-A1 and CE-B1, so PE1 needs to learn those routes from the CEs, redistribute them into BGP with some exported RT value, and advertise them to PE2 using IBGP, with PE2 then importing the correct routes (based on PE2's import RTs) into PE2's VRFs.
In fact, for simple VPN implementations, in which each VPN consists of all sites for a single customer, most configurations simply use a single RT value, with each VRF for a customer both importing and exporting that RT value.
MPLS can support overlapping VPNs by virtue of the RT concept. An overlapping VPN occurs when at least one CE site needs to be reachable by CEs in different VPNs.
Many variations of overlapping VPNs exist. An SP may provide services to many customers, so the SP actually implements CE sites that need to be reached by a subset of customers. Some SP customers may want connectivity to one of their partners through the MPLS network—for example, customer A may want some of its sites to be able to send packets to some of customer B's sites.
Regardless of the business goals, the RT concept allows an MPLS network to leak routes from multiple VPNs into a particular VRF. BGP supports the addition of multiple Extended Community PAs to each BGP table entry. By doing so, a single prefix can be exported with one RT that essentially means "make sure all VRFs in VPN-A have this route," while assigning another RT value to that same prefix—an RT that means "leak this route into the VRFs of some overlapping VPN."
Figure 19-17 shows an example of the concepts behind overlapping MPLS VPNs, in particular, a design called a central services VPN. As usual, all customer A sites can send packets to all other customer A sites, and all customer B sites can send packets to all other customer B sites. Also, none of the customer A sites can communicate with the customer B sites. However, in addition to these usual conventions, CE-A1 and CE-B2 can communicate with CE-Serv, which connects to a set of centralized servers.
Figure 19-17 Central Services VPN
To accomplish these design goals, each PE needs several VRFs, with several VRFs exporting and importing multiple RTs. For example, PE1 needs two VRFs to support customer A—one VRF that just imports routes for customer A, and a second VRF that imports customer A routes as well as routes to reach the central services VPN. Similarly, PE2 needs a VRF for the central services VPN, which needs to import some of the routes in VPN-A and VPN-B.
The MPLS VPN Data Plane
The explanations of the VRF, RD, and RT features explain most of the details of the MPLS VPN control plane. VRFs allow PEs to store routes learned from various CEs, even if the prefixes overlap. The RD allows PEs to advertise routes as unique prefixes, even if the IPv4 prefixes happen to overlap. Finally, the RT tells the PEs which routes should be added to each VRF, which provides greater control and the ability to allow sites to be reachable from multiple VPNs.
At the end of the process, however, to support the forwarding of packets, ingress PEs need appropriate FIB entries, with Ps and PEs needing appropriate LFIB entries. This section focuses on explaining how LSRs fill the FIB and LFIB when using MPLS VPNs.
As usual for this chapter, this section focuses on how to forward packets to subnet 10.3.3.0/24 in the customer A VPN. To begin this examination of the MPLS VPN data plane, consider Figure 19-18. This figure repeats the same forwarding example in Figure 19-13 but now shows a few details about the FIB in the ingress PE and the LFIB entries in the P and egress PE routers.
Figure 19-18 The Ingress PE FIB and Other Routers' LFIBs
The numbered steps in the figure are as follows:
- An unlabeled packet arrives on an interface assigned to VRF-A, which will cause ingress PE1 to use VRF-A's FIB to make a forwarding decision.
- Ingress PE1's VRF-A FIB entry for 10.3.3.0/24 lists an outgoing interface of S0/0/1, and a label stack with two labels—an inner label of 3333 and an outer label of 1111. So PE1 forwards the packet with these two labels pushed in front of the IP header.
- P1 uses the LFIB entry for incoming (local) label 1111, swapping this outer label value to 2222.
- PE2 does two LFIB lookups. PE2 finds label 2222 in the table and pops that label, leaving the inner label. Then PE2 looks up the inner label 3333 in the LFIB, noting the pop action as well, along with the outgoing interface. So PE2 forwards the unlabeled packet out interface S0/1/0.
The example shows the mechanics of what happens in the data plane once the correct FIB and LFIB entries have been added. The rest of this topic about the MPLS VPN data plane examines how MPLS VPN LSRs build these correct entries. While reading this section, it is helpful to keep in mind a couple of details about the purpose of the inner and outer label used for MPLS VPNs:
- The outer label identifies the segments of the LSP between the ingress PE and the egress PE, but it does not identify how the egress PE should forward the packet.
- The inner label identifies the egress PE's forwarding details, in particular the outgoing interface for the unlabeled packet.
Building the (Inner) VPN Label
The inner label identifies the outgoing interface out which the egress PE should forward the unlabeled packet. This inner label, called the VPN label, must be allocated for each route added to each customer VRF. More specifically, a customer CE will advertise routes to the PE, with the PE storing those routes in that customer's VRF. In order to prepare to forward packets to those customer subnets, the PE needs to allocate a new local label, associate the label with the prefix (and the route's next-hop IP address and outgoing interface), and store that information in the LFIB.
Figure 19-19 shows PE2's routes for 10.3.3.0/24 in both VRF-A and VRF-B and the resulting LFIB entries. The figure shows the results of PE2's process of allocating a local label for each of the two routes and then also advertising those labels using BGP. (Note that the LFIB is not a per-VRF table; the LFIB is the one and only LFIB for PE2.)
Figure 19-19 Creating the VPN Label LFIB Entry on the Egress PE
The steps shown in the figure are as follows:
- After adding a route for 10.3.3.0/24 to VRF-A, PE2 allocates a local label (3333) to associate with the route. PE2 then stores the local label and corresponding next hop and outgoing interface from VRF-A's route for 10.3.3.0/24 into the LIB (not shown) and LFIB.
- PE2 repeats the logic in Step 1 for each route in each VRF, including the route in VRF-B shown at Step 2. After learning a route for 10.3.3.0/24 in VRF-B, PE2 allocates a different label value (4444), associates that route's next-hop IP address and outgoing interface with the new label, and adds the information to a new LFIB entry.
- PE2 adds the local labels to the BGP table entry for the routes, respectively, when redistributing routes into BGP.
- PE2 uses IBGP to advertise the routes to PE1, with the BGP Update including the VPN label.
As a result of the first two steps in the figure, if PE3 receives a labeled packet and analyzes a label value of 3333, PE2 would be able to forward the packet correctly to CE-A2. Similarly, PE2 could correctly forward a received labeled packet with label 4444 to CE-B2.
Creating LFIB Entries to Forward Packets to the Egress PE
The outer label defines the LSP from the ingress PE to the egress PE. More specifically, it defines an LSP used to forward packets to the BGP next-hop address as advertised in BGP Updates. In concept, the ingress PE adds the outer label to make a request of the core of the MPLS network to "deliver this packet to the egress PE—which advertised this particular BGP next-hop address."
MPLS VPNs use an IGP and LDP to learn routes and labels, specifically to learn the label values to use in the outer label. To link the concepts together, it can be helpful to think of the full control plane process related to the LSP used for the outer label, particularly Step 4 onward:
- A PE, which will be an egress PE for this particular route, learns routes from some CE.
- The egress PE uses IBGP to advertise the routes to an ingress PE.
- The learned IBGP routes list some next-hop IP address.
- For MPLS VPNs to work, the PE and P routers must have advertised a route to reach the BGP next-hop addresses.
- Likewise, for MPLS VPNs to work, the PE and P routers must have advertised labels with LDP for the routes to reach the BGP next-hop addresses.
- Each P and PE router adds its part of the full end-to-end LSP into its LFIB, supporting the ingress PE's ability to send a packet to the egress PE.
For example, Figure 19-19 shows PE2 advertising two routes to PE1, both with BGP next-hop IP address 22.214.171.124. For MPLS to work, the collective PE and P routers need to advertise an IGP route to reach 126.96.36.199, with LDP advertising the labels, so that packets can be label switched toward the egress PE. Figure 19-20 shows the basic process; however, note that this part of the process works exactly like the simple IGP and LDP process shown for unicast IP forwarding in the first half of this chapter.
Figure 19-20 Creating the LFIB Entries to Reach the Egress PE's BGP Next Hop
The steps in the figure focus on the LFIB entries for prefix 188.8.131.52/32, which matches PE2's BGP next-hop IP address, as follows. Note that the figure does not show all LDP advertisements but only those that are particularly interesting to the example.
- PE2, upon learning a route for prefix 184.108.40.206/32, allocates a local label of 2222.
- PE2 updates its LFIB for the local label, listing a pop action.
- As normal, PE2 advertises to its LDP neighbors the label binding of prefix 220.127.116.11/32 with label 2222.
- P1 and P2 both independently learn about prefix 18.104.22.168/32 with the IGP, allocate a local label (1111 on P1 and 5555 on P2), and update their LFIBs.
- P1 and P2 advertise the binding of 22.214.171.124/32, along with their respective local labels, to their peers.
Figure 19-18 showed the FIB and LFIB entries required for forwarding a packet from CE-A1 to CE-A2, specifically into subnet 10.3.3.0/24. Figures 19-19 and 19-20, and their associated text, explained how all the LFIB entries were created. Next, the focus turns to the FIB entry required on PE1.
Creating VRF FIB Entries for the Ingress PE
The last part of the data plane analysis focuses on the ingress PE. In particular, the ingress PE uses the following logic when processing an incoming unlabeled packet:
- Process the incoming packet using the VRF associated with the incoming interface (statically configured).
- Forward the packet using that VRF's FIB.
The FIB entry needs to have two labels to support MPLS VPNs: an outer label that identifies the LSP with which to reach the egress PE, and an inner label that identifies the egress PE's LFIB entry that includes the correct outgoing interface on the egress PE. Although it might be obvious by now, for completeness, the ingress PE learns the outer and inner label values as follows:
- The outer label is based on the LIB entry, specifically for the LIB entry for the prefix that matches the BGP-learned next-hop IP address—not the packet's destination IP address.
- The inner label is based on the BGP table entry for the route in the VRF that matches the packet's destination address.
Figure 19-21 completes the ongoing example by showing the process by which PE1 adds the correct FIB entry into VRF-A for the 10.3.3.0/24 prefix. The figure picks up the story at the point at which PE1 has learned all required BGP and LDP information, and it is ready to populate the VRF routing table and FIB.
Figure 19-21 Creating the Ingress PE (PE1) FIB Entry for VRF-A
PE1's BGP table holds the VPN label (3333), while PE1's LIB holds the two labels learned from PE1's two LDP neighbors (P1 and P2, labels 2222 and 5555, respectively). In this case, PE1's best route that matches BGP next-hop 126.96.36.199 happens to point to P1 instead of P2, so this example uses label 1111, learned from P1.
The steps in the figure are explained as follows:
- PE1 redistributes the route from BGP into the VRF-A routing table (based on the import RT).
- PE1 builds a VRF-A FIB entry for the route just added to the VRF-A routing table.
- This new FIB entry needs to include the VPN-label, which PE1 finds in the associated BGP table entry.
- This new FIB entry also needs to include the outer label, the one used to reach the BGP next-hop IP address (188.8.131.52), so PE1 looks in the LIB for the best LIB entry that matches 184.108.40.206, and extracts the label (1111).
- Ingress PE1 inserts the MPLS header including the two-label label stack.
At this point, when PE1 receives a packet in an interface assigned to VRF-A, PE1 will look in the VRF-A FIB. If the packet is destined for an address in prefix 10.3.3.0/24, PE1 will match the entry shown in the figure, and PE1 will forward the packet out S0/0/1, with labels 1111 and 3333.
Penultimate Hop Popping
The operation of the MPLS VPN data plane works well, but the process on the egress PE can be a bit inefficient. The inefficiency relates to the fact that the egress PE must do two lookups in the LFIB after receiving the packet with two labels in the label stack. For example, the data plane forwarding example used throughout this chapter has been repeated in Figure 19-22, with a summary description of the processing logic on each router. Note that the egress PE (PE2) must consider two entries in its LFIB.
Figure 19-22 Two LFIB Lookups Required on the Egress PE
To avoid this extra work on the very last (ultimate) LSR, MPLS uses a feature called penultimate hop popping (PHP). (Penultimate simply means "1 less than the ultimate.") So the penultimate hop is not the very last LSR to process a labeled packet, but the second-to-last LSR to process a labeled packet. PHP causes the penultimate-hop LSR to pop the outer label, so that the last LSR—the ultimate hop if you will—receives a packet that only has the VPN label in it. With only this single label, the egress PE needs to look up only one entry in the LFIB. Figure 19-23 shows the revised data plane flow with PHP enabled.
Figure 19-23 Single LFIB Lookup on Egress PE Due to PHP