Definitive MPLS Network Designs [Electronic resources] نسخه متنی

Figure 2-19. Inter-AS MPLS Traffic Engineering

[INTER-AREA-AS]. Note that PCE-based TE LSP path computation techniques are discussed in the IETF PCE Working Group (see [PCE]). It relies on a router's ability to send a path-computation request to a Path Computation Element (PCE), which can be an ABR in the case of interarea MPLS TE or an ASBR for inter-AS MPLS TE. In a nutshell, the requesting router sends the request to a dynamically discovered (by means of IGP extensions) PCE attached to its local area or AS and referred to as the headend PCE. If the destination of the TE LSP belongs to an area the headend PCE is attached to, the PCE can compute the shortest path to the destination because it has the required information (indeed, the PCE has the network topology and resource information for both areas). Otherwise, the headend PCE relays the request to another PCE, which in turn computes all the shortest paths obeying the set of constraints from every ABR in the tail-end area to the TE LSP destination. Upon receiving the set of computed paths, the headend PCE can compute a shortest path end to end and return it to the requesting headend router. The PCE solution is recursive backward and relies on distributed path computation. It allows for the computation of end-to-end shortest paths and diverse paths if they exist. A detailed example is provided in Chapter 5 in the context of Inter-AS TE LSPs used to interconnect virtual POPs.

The next issue to consider is TE LSP reoptimization. Handling reoptimization in the context of interarea/AS MPLS TE is slightly more challenging than in the case of a flat IGP area. Indeed, how does the headend router detect that a more optimal (shortest) path exists if it happens to be in another area/AS?

One approach might be to systematically try to resignal any interarea/AS TE LSP without knowing beforehand whether a more optimal path exists. If it turns out that one of the ABRs can find a better path, the TE LSP follows the better path. However, such a solution has the drawback of systematically resignaling the interarea/AS TE LSP (which implies RSVP processing, new label allocation, and so forth) even if no better path exists.

Hence, a solution (available on Cisco routers) has been proposed in [LOOSE-PATH-REOPT]. It allows the headend to be informed of the existence of a more optimal path in some downstream area/AS, either after having explicitly sent an inbound request or by means of explicit notification from the ABR that has detected the presence of a better path. In the former case, the headend router simply sets a specific bit in the RSVP Path message that triggers the reevaluation of the existing path on each node whose next hop is defined as a loose hop (for instance, ABRs 1 and 3 in [INTER-AS-TE-REQS]. In this section, we will just highlight the most significant specificities of inter-AS MPLS TE.

The first specificity of inter-AS TE is related to the existence and nonvisibility of the inter-ASBR links. In most situations, autonomous systems are interconnected by non-IGP-enabled inter-ASBR links (the routing protocol in use is usually BGP-4). Hence, a headend router does not have the visibility of the ASBR links and in particular the TE link characteristics such as bandwidth, affinities, and so on.

One obvious solution is to use the per-AS path computation approach described earlier and specify all the ASBRs as loose hops. However, in the case of an inter-AS TE LSP, this requires configuring quite a significant number of hops (such as ASBR1, ASBR3, ASBR5, ASBR7, and R9 for an inter-AS TE LSP from R1 to R9 in Figure 2-20). You also have to configure a potentially large number of combinations of loose hops on the headend router to handle the case of LSP setup failure. Furthermore, having the visibility of the inter-ASBR would allow the headend to make a more efficient path computation.

For example, suppose that the link ASBR1ASBR3 does not meet the bandwidth requirement of T1. Because R1 (the headend in this example) does not know the state of that link, it may try to signal the TE LSP across that link. In this case, ASBR1 would reject the TE LSP (by returning an RSVP Path Error). R1 would then try another combination of loose hops. Hence, a more efficient solution implemented on Cisco routers consists of flooding inter-ASBR link state, although no IGP routing adjacency is established between the two ASBRs. This allows a reduction in the number of ASBRs that are required to be configured as loose hops and improves the path computation efficiency. This is illustrated in Figure 2-21.