PCE BOF JP Vasseur (Ed.) Internet Draft Anna Charny Francois Le Faucheur Cisco System Inc. Javier Achirica Telefonica JL Le Roux (Ed.) France Telecom Category: Standard Track Expires: January 2005 July 2004 Framework for PCE-based MPLS-TE Fast Reroute Backup Path Computation draft-leroux-pce-backup-comp-frwk-00.txt Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet- Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt. The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. By submitting this Internet-Draft, I certify that any applicable patent or IPR claims of which I am aware have been disclosed, or will be disclosed, and any of which I become aware will be disclosed, in accordance with RFC 3668. Vasseur, Le Roux et al. Expires January 2005 [Page 1] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 Abstract This document proposes a framework for the use of Path Computation Elements (PCE) to compute bypass tunnels paths, in the context of the MPLS-TE Fast Reroute, while allowing bandwidth sharing between bypass tunnels protecting independent resources. Both a centralized and a distributed PCE scenarios are described. The corresponding required Routing and signalling extensions are beyond the scope of this framework and will be addressed in separate documents. Conventions used in this document The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC-2119. 0. Background This draft is a new revision of the expired draft draft-vasseur-mpls- backup-computation-02. Table of Contents 1. Terminology.................................................4 2. Introduction................................................5 3. Requirements................................................5 3.1. Bandwidth Protection........................................5 3.2. Bandwidth sharing...........................................6 4. Bypass tunnel path computation models.......................7 5. Limitations of the independent CSPF-based computation model.8 5.1. Bandwidth sharing...........................................8 5.2. Potential inability to find a placement of a set of bypass.tunnels satisfying constraints.......................9 6. The PCE-based Computation Model.............................9 6.1. Solution Overview...........................................9 6.2. Information required by the PCE............................10 6.3. Centralized PCE scenario...................................11 6.3.1. PCE responsible for both the primary and bypass tunnels path.computation...........................................11 6.3.1.1. PLR-PCE Signaling........................................12 6.3.1.2. Signaling Bypass tunnels with zero Bandwidth.............12 6.3.2. PCE responsible for bypass tunnels path computation only (not primary TE LSPs)......................................13 6.3.2.1. Backup Pool advertised in IGP............................13 6.3.2.2. Backup Pool not being advertised in IGP..................14 Vasseur, Le Roux et al. Expires December 2004 [Page 2] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 6.4. Distributed PCE scenario...................................14 6.4.1. Node Protection............................................14 6.4.2. Link Protection............................................16 6.4.3. SRLG protection............................................16 7. Validity of the Independent failure assumption.............17 8. Operations with links belonging to multiple SRLGs..........19 8.1. Notion of SRLG dependency, and Shared SRLG Dependency Link Group (SDLG)..........................................20 8.2. SDLG protection............................................21 8.2.1. Distributed scenario for SDLGs protection..................22 8.3. Alternative solution.......................................22 9. Operations with DS-TE and multiple Class-Types.............22 9.1. Single backup pool.........................................23 9.2. Multiple backup pools......................................25 10. Interaction with scheduling................................27 11. PLR-PCE Signaling: Functional description..................28 11.1. Element to protect.........................................28 11.2. Bandwidth to protect.......................................28 11.3. Affinities.................................................28 11.4. Maximum number of bypass tunnels...........................28 11.5. Minimum bandwidth on any element of a set of bypass tunnels....................................................28 11.6. Class Type to protect......................................29 11.7. Set of already in place bypass tunnels.....................29 12. Bypass tunnel - Make before break..........................29 13. Stateless versus Statefull PCE.............................29 14. Packing algorithm..........................................29 15. Security Considerations....................................30 16. Acknowledgements...........................................30 17. Security Considerations....................................30 18. Intellectual Property Statement............................30 18.1. IPR Disclosure Acknowledgement.............................30 19. References.................................................31 20. Authors' Address:..........................................32 Vasseur, Le Roux et al. Expires December 2004 [Page 3] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 1. Terminology Terminology used in this document LSR - Label Switch Router LSP - An MPLS Label Switched Path PCE - Path Computation Element. A PCE may be any kind of LSR or an offline tool not forwarding packets. A PCE computes paths for TE-LSPs it is not the head-end for. PCC - Path Computation Client (any LSR) requesting a path computation of the Path Computation Element. Local Repair - Techniques used to repair LSP tunnels quickly when a node or link along the LSPs path fails. Protected LSP - An LSP is said to be protected at a given hop if it has one or multiple associated bypass tunnels originating at that hop. Bypass Tunnel - An LSP that is used to protect a set of LSPs passing over a common facility. PLR - Point of Local Repair. The head-end of a bypass tunnel. MP - Merge Point. The LSR where bypass tunnels meet the protected LSP. A MP may also be a PLR. NHOP Bypass Tunnel - Next-Hop Bypass Tunnel. A bypass tunnel which bypasses a single link of the protected LSP. NNHOP Bypass Tunnel - Next-Next-Hop Bypass Tunnel. A bypass tunnel which bypasses a single node of the protected LSP. Reroutable LSP - Any LSP for which the "Local protection desired" bit is set in the Flag field of the SESSION-ATTRIBUTE object of its Path messages (and/or a FAST-REROUTE object is included in its Path message). CSPF - Constraint-based Shortest Path First. Vasseur, Le Roux et al. Expires December 2004 [Page 4] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 2. Introduction This document proposes a framework for the use of Path Computation Element(s) to compute bypass tunnels paths, in the context of the MPLS-TE Fast Reroute, while allowing bandwidth sharing between bypass tunnels protecting independent resources. Both a centralized and a distributed PCE scenarios are described. The focus of this document is ''Bandwidth protection'' in the context of local repair capability of MPLS Fast Reroute. Bandwidth Protection refers to the issues related to the computation of a bypass tunnel path satisfying the capacity requirements of the primary tunnels protected by the considered bypass tunnel. We do not propose another method for MPLS Traffic Engineering Fast Reroute. This draft makes the assumption that the fast reroute technique named Facility backup and described in [FAST-REROUTE] is used to provide fast recovery in case of link/node failure. The exact algorithms for placement of the bypass tunnels with bandwidth guarantees are outside the scope of this draft. Rather, we concentrate on the mechanisms enabling the bypass tunnel path computation to be performed by a Path Computation Element (PCE) which holds sufficient information in order to achieve efficient sharing of bandwidth between bypass tunnels protecting independent failures. The mechanisms are described in the context of both a centralized (the PCE computes the set of bypass tunnels to protect every facility in the network) and a distributed computation (every LSR behaves a PCE to compute the set of bypass tunnels for each of its neighbours in case of its own failure or failure of one of its own links). The required routing and signalling extensions (PLR-PCE signalling) will be addressed in separate documents. 3. Requirements 3.1. Bandwidth Protection As defined in [FAST-REROUTE], a TE LSP can explicitly request to be fast protected (in case of link/node/SRLG failure the TE LSP will be locally rerouted onto a backup tunnel, as defined in [FAST REROUTE]) and rerouted onto a backup tunnel with an equivalent bandwidth (in other words without QOS degradation, supposing here that offering an equivalent QOS can be reduced to preserving bandwidth requirement). This can be signaled (in the Path message) in two ways: - with the SESSION-ATTRIBUTE object by setting: - the ''Local protection desired'' bit - the ''Bandwidth protection desired'' bit - with the FAST-REROUTE object by setting a non-zero bandwidth. Note that other parameters related to the backup tunnel can also be signaled in the Path message. Vasseur, Le Roux et al. Expires December 2004 [Page 5] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 Bandwidth protection would typically be requested for TE LSPs carrying very sensitive traffic (Voice trunking, ...). When a link/node/SRLG failure occurs, the PLR (Point of Local Repair) fast reroutes the protected LSPs onto their bypass tunnel. The PLR should also send a Path Error notifying the head-end LSRs that the protected LSPs have been locally repaired so that head-ends can trigger a re-optimization, and potentially reroute the TE LSP in a non disruptive manner (make before break) following a more optimal path, provided such a path exists. The bandwidth of the bypass tunnels that the protected LSPs will be rerouted onto will dictate the level of bandwidth protection and in turn the QOS during failure until the TE LSPs are being re-optimized by their respective head-end (if such a re-optimization can be performed, depending on the available network resources). Various constraints can be taken into account for the bypass tunnels: (1) must be diversely routed from the protected element (link/node/SRLG diverse), (2) must be setup in such a way that they get enough bandwidth so that the protected LSPs requesting bandwidth protection should receive the same level of QOS when rerouted. Note that the notion of bandwidth protection is on a per LSP basis. (1) must always be satisfied and makes FRR an efficient protection mechanism to reroute protected TE LSP in 10s of milliseconds in case of link or node failure. (2) allows FRR to provide an equivalent level of QOS during failure to the TE LSPs that have requested bandwidth protection. A backup Path computation mechanism ensuring bandwidth protection is highly desirable. 3.2. Bandwidth sharing Since local repair is expected to be used for only a short period of time after failure, typically followed by re-optimization of the affected primary LSPs, it is reasonable to expect that the probability of multiple failures (of facilities which do not share an SRLG) in this short period of time is very small. As a result, being able to share bandwidth on the link among bypass tunnels protecting independent facilities (with regards to failure risks) typically results in large savings in the bandwidth required for protection. This is what we refer many times in this document as ''efficient bandwidth sharing'' or as achieving ''bandwidth sharing''. Note also that the single failure assumption needed for such bandwidth sharing is a pre-requisite to any protection approach which uses pre-computed protected paths, clearly even two completely link and node disjoint Vasseur, Le Roux et al. Expires December 2004 [Page 6] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 pre-computed paths can both fail if more than one failure can occur as one failure may occur on the primary and the other on the second path. It is worth underlining that the single failure of a SRLG may result in the actual failure of multiple links. For the purposes of this draft we consider the entire SRLG as a single element that needs to be protected. Once the head-end receives the Path Error (''Tunnel locally repaired''), reoptimization should be triggered followed by an LSP reroute making use of the ''Make Before Break'' technique to avoid traffic disruption, assuming such a more optimal path obeying the constraints within the new network topology can be found. If such a path cannot be found, the TE LSP will not be reoptimized and will still be fast rerouted by the immediately upstream PLR attached to the failed element. The two following situations result in a multiple independent failures scenario where bandwidth protection with backup bandwidth sharing cannot be ensured: - a second failure occurs before the TE LSP is reoptimized, - the TE LSP cannot be reoptimized and a second failure happens before the first failure has been restored. Note however that in networks where bandwidth is a reasonably available resource, this situation is unlikely to happen as the TE LSP reoptimization will succeed. Furthermore, in networks where bandwidth is a very scarce resource, bandwidth protection without backup bandwidth sharing is likely to require substantially more bandwidth, and therefore is likely to be impossible anyway. As a result, bandwidth sharing among bypass tunnels protecting independent failures is highly desirable. 4. Bypass tunnel path computation models Various bypass tunnel path computation models have been proposed: independent CSPF-based computation, [KINI], [BP-PLACEMENT], ... A new model, named ''PCE based computation model'' is proposed in this draft. Vasseur, Le Roux et al. Expires December 2004 [Page 7] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 5. Limitations of the independent CSPF-based computation model The simplest mechanism (called independent CSPF-based computation model) to get bandwidth protection available is to rely on existing IGP TE advertisement and for the head-end of the bypass tunnel: - to determine the bandwidth requirements of the desired bypass tunnel(s), - to compute the bypass tunnels path in the network where the appropriate amount of bandwidth is available using standard CSPF-based computation, - to signal the bandwidth requirements of the individual bypass tunnels explicitly. While this approach is quite attractive for its simplicity, it presents a substantial set of challenges: - Inability to perform bandwidth sharing between bypass tunnels protecting independent resources, resulting in very large wastage of bandwidth. - Potential inability to find a placement of the bypass tunnels satisfying the bandwidth constraints, even if such a placement exist 5.1. Bandwidth sharing Improvements to the independent CSPF-based computation model have been proposed in [KINI] and [BP-PLACEMENT] in order to achieve bandwidth sharing. They still rely on an independent CSPF computation performed by PLRs. In [BP-PLACEMENT], routing extensions are proposed to propagate the set of bypass LSPs and their attributes, allowing for a complete bandwidth sharing, but with a significant impact on the IGP. While the approach described in [KINI] substantially reduces the amount of information that needs to be propagated in routing updates, it sacrifices the amount of achievable sharing. Both approaches also require modifications to admission control algorithms, as well as signaling extensions to convey the information necessary to perform specific call admission control for backup LSPs. In contrast, the approach described in this draft can be used to achieve complete sharing without any routing extensions and without any modification to admission control (although as discussed further in section 7.2 a small amount of routing extensions is desirable for the distributed case to provide flexibility in the choice of protection strategies) Vasseur, Le Roux et al. Expires December 2004 [Page 8] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 5.2. Potential inability to find a placement of a set of bypass tunnels satisfying constraints Another well-known issue with independent CSPF-based computation with explicitly signaled bandwidth requirements is its potential inability to find a placement of the bypass tunnels satisfying the bandwidth constraints, even if such a placement exists. This issue is not specific to the placement of the bypass tunnels - rather it is due to the sub-optimality of a greedy on-demand nature of the CSPF approach and the non coordinated bypass tunnel computation approach to protect a given facility While addressing this problem is not a primary goal of this draft, The PCE-based computation model described in this draft provides the opportunity to improve the chance of finding a feasible placement of the bypass tunnel as it enables the use of algorithms that can take advantage of coordination between the placement of bypass tunnels protecting the same element. However, the exact algorithms appropriate for this purpose are outside of the scope of this draft. 6. The PCE-based Computation Model This draft proposes another model for bypass tunnel path Computation, based on the use of PCE capabilities. It is referred as the ''PCE based computation model''. Note that in this section we assume that a bypass LSP protects only one facility (link, node or SRLG). The PCE based computation model can be extended to more general case where bypass tunnel can protect more than one facility, but this requires specific procedures that are addressed in sections 7 (NNHOP activated in case of both link and node failures) and 8 (NHOP protecting link belonging to multiple SRLGs). 6.1. Solution Overview The proposed solution consists in moving the bypass computation from the PLR to a Path Computation Element that computes, in a coordinated manner, all bypass LSPs protecting a given facility. Given the single failure assumption, the set of bypass tunnels protecting a given facility is computed independently of any other bypass tunnel protecting a distinct facility, and making use of all bandwidth available for backup purpose. This directly ensures bandwidth sharing. Also, the sets of bypass tunnels protecting independent facilities can be computed by distinct PCEs, allowing for a distributed scenario. As bypass tunnels protecting a common facility are computed in a coordinated manner and as they do not compete for bandwidth with Vasseur, Le Roux et al. Expires December 2004 [Page 9] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 bypass tunnels protecting independent facilities, they can be signalled with zero bandwidth. This avoids any extension to RSVP and local admission control, to take into account bandwidth sharing. This requires, when primary and bypass tunnel are not computed by the same entity, the definition of a backup-bandwidth pool because the bandwidth used by the bypass tunnels is invisible to the entity responsible for the computation of the primary TE LSPs. The PCE based computation model can be implemented in two different ways: centralized or distributed. In the centralized case a single PCE computes the protection of all facilities. We distinguish two options whether the PCE is also responsible for the placement of primary tunnels or not. In the distributed case the PCE function is distributed on LSRs. The distribution should be so that all bypass tunnels protecting a given facility are computed by the same LSR: -For node protection, the PCE is the protected LSR itself -For link protection, the PCE is one of the two link end-points -For SRLG protection, the PCE is one elected LSR among the end- points of the SRLG links end-points. In all of these scenarios the PCE-based computation enables sharing of bandwidth among bypass tunnels protecting independent failures. In addition, all of these scenarios also allow overcoming some of the limitations of the greedy independent CSPF-based placement of the bypass tunnels, increasing the chances of finding a bypass tunnels placement satisfying the constraints if such a solution exists. While some of these approaches can benefit from an IGP-TE extension advertising an additional backup bandwidth pool, all of these approaches can be usefully deployed in a limited fashion in the existing networks without any additional routing extensions at all. Some signaling protocol is required to allow communication between PLRs and PCE, so that the PLR can request the PCE for backup computation and the PCE can reply with the computed paths. The required routing and signaling protocol extensions will be addressed in separate documents. 6.2. Information required by the PCE To compute the bypass tunnels protecting a given element, the PCE needs to know: - the network topology, - the desired amount of primary traffic that needs to be bandwidth protected (this could be either the actual bandwidth reserved by primary TE LSPs requiring bandwidth protection or the whole bandwidth pool from which the primary LSPs reserve bandwidth) - the amount of bandwidth available for the placement of the bypass tunnels (also referred to as backup bandwidth). Vasseur, Le Roux et al. Expires December 2004 [Page 10] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 The network topology is available directly from the IGP TE database as well as the desired amount of primary traffic that needs to be protected if one protects a bandwidth pool (and not the actual bandwidth reserved by primary TE LSPs requiring bandwidth protection). The information about the backup bandwidth pool depends on the exact model and is discussed separately in above scenarios. 6.3. Centralized PCE scenario In the centralized scenario, the bypass tunnel path computation is being performed on a central PCE (which can be a workstation or another LSR). The PCE will be responsible for the computation of the bypass tunnels for some or all the LSRs in the network. Typically, there could be one PCE per area in the context of a multi-area network. The PCE(s) address(es) may be manually configured on every LSR or automatically discovered using IGP extensions. WE distinguish two cases, whether the PCE is also responsible for the computation of primary tunnels or not. There differ in the way backup bandwidth is handled. 6.3.1. PCE responsible for both the primary and bypass tunnels path computation In this scenario, the PCE can easily take advantage of knowing all the primary tunnels to define bandwidth protection requirements based on actual primary LSPs. There is substantial flexibility in choosing what bandwidth can be used for the bypass tunnel placement. One approach might be to use for the bypass tunnels the same bandwidth pool as the corresponding primary LSPs. At some point the user will have to specify the policy to the server. For example, protect traffic of a pool X with a bypass tunnel in the same pool but also the proportion of pool X that can be used for backup and primary. For pool X, the user could specify: ''up to y% of pool X can be used for backup''. Since in this scenario the server is responsible for the placement of both the primary traffic and the bypass tunnels, at any given time in the computation of the bypass tunnels it has complete information about the topology and the current placement of all bypass and primary tunnels. Therefore, the server can compute the bypass tunnels protecting one element at a time, and when placing its bypass tunnels simply ignore the bandwidth of any bypass tunnels already placed if Vasseur, Le Roux et al. Expires December 2004 [Page 11] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 those protect a different element, thus ensuring implicitly the desired bandwidth sharing. In this case, there is no need to specify a notion of backup bandwidth pool. 6.3.1.1. PLR-PCE Signaling A PLR has to request the PCE for bypass path computation, potentially indicating the amount of bandwidth that has to be protected. Having computed the bypass tunnels, the PCE needs to inform the PLR, about the placement of the bypass tunnels, their bandwidth, and the elements they protect. Such PLR-PCE communication requires an LSR-PCE signalling protocol for communication of bypass path computation request from the LSR to the PCE and bypass path computation responses from the PCE to the LSR. 6.3.1.2. Signaling Bypass tunnels with zero Bandwidth Once an LSR has received the information about the bypass tunnels for one or more elements it is the head-end for, it needs to establish those tunnels along the specified paths. At first glance, given the need to ensure bandwidth protection, it seems natural to signal the bandwidth requirements of the bypass tunnel explicitly. However, as discussed in [BP-PLACEMENT], such approach requires that the local admission control is changed to be aware of the bandwidth sharing, and additional signaling extensions need to be implemented to enable an LSR to tell a primary LSP from a bypass LSP so that admission control can be performed differently in the two cases. However, since the placement of both the primary and the bypass tunnels in this case is done by the server which maintains the bandwidth requirements of all these primary and bypass LSPs, it is sufficient to signal zero-bandwidth tunnels, thus avoiding the need for any additional signaling extensions or changes to admission control. Even though the required bandwidth will not be explicitly signaled, it will nevertheless be available along the path upon failure by virtue of the computation of this placement by the server which is fully aware of the global topology and places all TE LSPs in such a way that their bandwidth requirements are satisfied. Note also that although the bandwidth requirements are not explicitly signaled, the PLR may store this information locally, since it may be needed in determination of which primary LSPs to assign to which bypass tunnels in the case where more than one bypass tunnel exists (see section 14). Vasseur, Le Roux et al. Expires December 2004 [Page 12] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 6.3.2. PCE responsible for bypass tunnels path computation only (not primary TE LSPs) A key aspect of the previous scenario (PCE computing both the primary and backup LSPs) was the fact that the PCE could make use, for the bypass tunnels, of any available bandwidth not reserved for primary TE LSPs. As a consequence, definition of a separate backup pool was not required. On the other hand, if the PCE is just responsible for the bypass tunnels paths (i.e the primary tunnels are established on-line or by any other mechanism external to the backup path computation server), and if the bypass tunnels are signaled with zero bandwidth to enable efficient bandwidth sharing, then the bypass tunnels cannot draw bandwidth from the same pool as the primary traffic they protect. This is because the bandwidth used by the bypass tunnels is invisible to the entity responsible for the computation of the primary TE LSPs and therefore the primary TE LSPs could make use of the entire bandwidth of a given pool. Therefore if the PCE used for bypass tunnel path computation uses any bandwidth of the same pool bandwidth protection violation might occur. Achieving efficient bandwidth sharing in this case requires the definition of a separate pool that could only be used for bypass tunnels. We refer to this pool as a backup pool. As in the previous approach: - A Signaling protocol is required for communication between PLRs and the PCE. - Bypass tunnels are signalled with zero bandwidth Note that the notion of backup bandwidth pool is similar to that described in [BP-PLACEMENT]. The backup bandwidth pool approach can be used in two ways: - being advertised in IGP - without being advertised in IGP 6.3.2.1. Backup Pool advertised in IGP In this approach, an additional bandwidth pool is established, and is flooded in the routing updates. If the backup PCE uses the value of the backup bandwidth pool for its computation, no bandwidth overbooking will ever occur, since the primary tunnels now use the bandwidth from a different pool. The additional state that needs to be flooded in routing updates to implement the backup bandwidth pool does not impact the IGP scalability as the bandwidth protection pool being announced by IGP-TE is a static value, it does not dynamically change as backup TE LSP are set up, which preserves IGP scalability. As the bandwidth protection pool is being defined on a per link basis, this allows for different policies depending on the link characteristics. Vasseur, Le Roux et al. Expires December 2004 [Page 13] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 6.3.2.2. Backup Pool not being advertised in IGP The routing extensions discussed in the previous section are desirable but not necessary to deploy this approach in the existing network in a limited, but nevertheless useful fashion. Since the computation of the bypass tunnels in this approach is performed by a centralized PCE, the PCE can use the notion of the backup bandwidth pool implicitly. Just as in the case of a PCE computing the placement of both primary and backup LSPs, such policy may be simply configured on the PCE for every link. The policy must ensure that the backup pool never overlaps with the pool requiring bandwidth protection. A generic approach could be for the PCE to compute, for each link, the backup bandwidth as: link-bandwidth - maximum-reservable- bandwidth. This approach requires that link-bandwidth > maximum- reservable-bandwidth which prevents the user from allowing TE overbooking. 6.4. Distributed PCE scenario While there are several clear advantages of a centralized (off-line) model, there are also well-known disadvantages of it (such as the single point of failure, the necessity to provide reliable communication channels to the server, etc.) While most of these issues can be addressed by the proper architectural design of the network, a dynamic distributed solution is clearly desirable. This section presents the use of the ''facility-based computation'' solution in a distributed bypass path computation scenario. 6.4.1. Node Protection Consider first the problem of node protection. The key idea is to shift the computation of the bypass tunnels from the head-ends of those bypass tunnels to the node that is being protected. Essentially, each node protects itself by computing the placement of all the bypass tunnels that are required to protect the bandwidth of traffic traversing this node in the case of its failure. Once the bypass tunnels are computed, they need to be communicated to their head-ends (in this case the neighbors of the protected node) for installation. The bypass tunnel head-ends play the role of PLR. Essentially, each node becomes a PCE for all of its neighbors, computing all NNHOP bypass tunnels between each pair of its neighbors which are necessary for its own protection. The fact that the bypass tunnels to protect a node X are being computed by a single PCE (node X) is essential and much more efficient than the non-coordinated independent CSPF-based computation. Vasseur, Le Roux et al. Expires December 2004 [Page 14] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 The key pieces that make this model work are those already described in the context of the centralized PCE: 1) Making use of explicitly defined backup bandwidth pool which is logically disjoint from the primary bandwidth pool, 2) Taking advantage of a single failure assumption to achieve bandwidth sharing, 3) Installing bypass tunnels with zero bandwidth. These three things together allow the computation of the placement of bypass tunnels for a given node to be completely independent of the placement of bypass tunnels for any other node. Essentially, each node has the entire backup bandwidth pool available for itself. The problem it needs to solve is how to place a set of NNHOP bypass tunnels (one or more for each pair of its direct neighbors) in a network with available capacity on each link equal to the backup bandwidth pool. This problem can be solved by any algorithm for finding a feasible placement of a set of flows with given demands in a network with links of given capacity. While the details of such algorithm are beyond the scope of this draft, it is clear that since the node now has control over all bypass tunnels protecting itself, it is more likely that it can find such a placement, and potentially find a more optimal placement, than is possible if the PLRs compute the placement of these tunnels independently of each other. Just as in the case of a centralized PCE, installing the bypass tunnels with zero bandwidth ensures that no changes to admission control are necessary to allow sharing of the backup pool by bypass tunnels protecting different nodes, thus enabling bandwidth sharing between independent failures. Yet, by virtue of the computation, the bypass tunnels protecting a given node will also have enough bandwidth in the case of that node's failure. Note also that the backup pools can be implicitly derived from the routing information already available. This could be done by configuring max global reservable pool to being less than the link speed by the desired value of the backup pool. Every node computing its bypass tunnels then can by default use link speed minus the max global reservable pool as the value of the backup pool to use in its computation of the bypass tunnels placement. As described earlier, there is substantial benefit in defining the backup pool explicitly and advertise its value as part of the topology in the routing updates. This clearly requires an IGP-TE extension that will be defined in a separate draft. The benefit of Vasseur, Le Roux et al. Expires December 2004 [Page 15] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 doing so is that it provides much more flexibility in the design of the network. Yet it is important to emphasize that while IGP-TE extensions is a clear benefit for facility-based computation, it is not a requirement for this solution to work under a limited set of assumptions (namely, as discussed above if the backup pool is set to link speed minus maximum reservable primary bandwidth, the latter being configured to less than link speed). Finally, as for a centralized PCE, a signaling protocol is required for communication between the protected node acting as PCE and its neighbour acting as PLRs. 6.4.2. Link Protection In order to protect a link with MPLS TE Fast Reroute in both directions, two bypass tunnels protecting each direction of this link are installed by the corresponding head-end of that link. To make sure that traffic requesting bandwidth protection traversing this link is protected in case of a link failure (if both directions fail simultaneously), it is necessary to account for the interaction of the bypass tunnels protecting different directions of this link. That is, one needs to make sure that if a bypass tunnel T1 protecting bandwidth B1 on a directed link A->B and the tunnel T2 protecting bandwidth B2 on a directed link B->A traverse the same directed link L, then link L has spare capacity of at least B1+B2. If the two ends of the link compute their bypass tunnels independently, the way to ensure this condition would be to explicitly signal the bandwidth of the bypass tunnels. However, as discussed earlier, this approach makes the sharing of bandwidth between the bypass tunnels protecting different elements impractical and would require IGP and admission control extensions. To achieve this goal in a distributed setting we propose that one of the two end-nodes of the link takes the responsibility for computing the bypass tunnels for both directions using the backup pools explicitly or implicitly defined. We propose that by default the node with the smaller IGP id serves as the PCE for the other end of the link. Therefore, by default a node with id X serves as a PCE for NNHOP bypass tunnels protecting itself and NHOP bypass tunnels protecting any adjacent bi-directional link for which the other end has an IGP id larger than X. 6.4.3. SRLG protection In the case when each link in the network cannot belong to more than one SRLG, we propose to use exactly the same approach as for the bi- directional link. That is, if an SRLG consists of a set of bi- directional links, the node with the smallest IGP id of all the Vasseur, Le Roux et al. Expires December 2004 [Page 16] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 endpoints of these links serves by default as a PCE. The case where links are part of more than one SRLG requires specific processing (see section 8). 7. Validity of the Independent failure assumption The facility based computation model is heavily dependent on the single independent failure assumption. That is, it is assumed that the probability of multiple independent element failures in the interval of time required for the network to re-optimize primary tunnels affected by a given failure and to re-compute the bypass tunnels for other elements is low. In a distributed model both of these tasks are likely to be accomplished within a very short time so the assumption typically can be justified. The loss of bandwidth protection in the rare cases that the assumption is violated is offset by the benefit of sharing the bandwidth between bypass tunnels protecting different elements. However, not all elements are independent. One example of elements that are not independent is a set of links in the same SRLG. Therefore, as discussed above, SRLG is treated as a single element and is protected as a single entity. Another example of failures that are not independent is a failure of a node and links adjacent to it. It is possible (and is frequently the case) that a failure of a node results also in the failure of the link(s). However, in the approach described in the draft the computation of bypass tunnel paths for link and node protection is done independently. This is necessary to ensure that NNHOP tunnels for a node can be computed completely independently of the NHOP tunnels for adjacent links, thus enabling the distributed computation. The justification for this is that when a node fails, traffic that does not terminate at this node is protected because it is rerouted over the NNHOP tunnels, and traffic that does terminate at the failed node does not need to be protected against the failure of adjacent links since it would get dropped anyway. Thus, the underlying assumption is that if a node fails, the NHOP tunnels protecting the link are not used, while if a link fails but the router does not, the NHOP tunnels are used. So they can in fact be computed independently. However, this reasoning only works if it is in fact possible to identify the type of failure correctly and use the appropriate set of tunnels depending on the failure. There are several cases to be considered: - A downstream router fails but the link does not, - The link fails but the downstream router does not, - The link fails because the downstream router failed. Vasseur, Le Roux et al. Expires December 2004 [Page 17] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 The first case is typically identifiable by means of RSVP hello or some IGP hellos mechanism on layer 2 link providing fast failure notification. However, when a link failure does occur, using the currently deployed mechanisms, a node adjacent to the failed link cannot tell within the time appropriate for Fast Reroute whether the node on the other side of that link is operational or not. Therefore, it is currently impossible to reliably tell apart the second and the third cases above. Hence, to protect both traffic that terminates at the failed node in case the failure was a link failure, and at the same time to protect traffic transit through the failed node in case it was a node failure, the LSR adjacent to the failed link is forced to use both the NHOP and the NNHOP tunnels at the same time. This may lead to a violation of bandwidth guarantees, since the NHOP and NNHOP tunnels were computed independently using the same backup bandwidth pool, and so they may share a link with enough bandwidth for only one but not the other. A similar issue occurs in the case of bi-directional link failure. Since the two nodes on each side of the link will see the failure of an adjacent link, unless they can detect that it was a link and not a node failure, they will be forced to activate the NHOP tunnel protecting the link, and the NNHOP tunnel protecting the node on the other side. Essentially, the system will operate as if two failures have occurred simultaneously when in reality only a single (bi- directional) link failed. This clearly can result in a violation of a bandwidth guarantee. To address this issue, one needs a mechanism to differentiate a link from a node failure. Such a mechanism is described in [LINKNODE- FAILURE]. Note that in the centralized model, the PCE may compensate for the lack of the ability to tell a link from a node failure by making sure that the NNHOP bypass tunnels for adjacent nodes and the NHOP bypass tunnels for the corresponding links do not collide. While this makes the problem of finding such backup tunnels algorithmically more challenging, it remains possible to achieve bandwidth sharing in this case. However, the ability to tell a link from a node failure is crucial for the distributed model when node protection is desired. It is worth mentioning however that if just NHOP bypass tunnels are required (nodes are considered as reliable ''enough'') and just links are protected against failures, then there is no need to distinguish between node and link failure even in the distributed case. Vasseur, Le Roux et al. Expires December 2004 [Page 18] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 8. Operations with links belonging to multiple SRLGs In section 7 we limit the study to the case of links that are not part of more than one SRLG. However in some networks links might be part of more than one SRLG. This section presents the use of the PCE based computation model in the general case where links are part of zero, one or more SRLGs. Both centralized and distributed scenarios are addressed. Recall that PCE based computation model consists of a coordinated placement of the set of bypass protecting one element by the same PCE, independently of the protection of each other element. This is clearly not applicable when bypass tunnels protect multiple independent elements, which is the case when bypass tunnels protect links belonging to multiple SRLGs, as an SRLG can be considered as an independent element (in terms of failure risk). In case SRLGs are not disjoint, the placement of bypass LSPs protecting a given SRLG cannot be done independently of any other SRLG. Even if SRLGs remain independent elements in term of failure risk, their bandwidth protection computation can no longer be done independently, and must be coordinated. For instance, lets take 3 links L1, L2, L3 and two SRLGs S1 and S2 such that S1= {L1, L2} and S2={L2, L3}. S1 and S2 are not disjoint, and their intersection is the link L2. If b1, b2 and b3 are NHOP bypass tunnels protecting respectively L1, L2, and L3 then: - b1 and b2 computations must be coordinated, as they protect a common SRLG S1. - b2 and b3 computations must be coordinated as they protect a common SRLG S2. It results clearly that b1, b2 and b3 path computations must be coordinated, (and thus in the framework of facility-based computation model must be performed by the same PCS) and we say that L1, L2 and L3 are SRLG dependant. It is important to note in this case that even if b1 and b3 protect independent elements, in terms of failure (L1 and L3 are SRLG diverse), their path computation must be coordinated. Bandwidth sharing can still be ensured in that case, but this additional level of dependency in the computation of bypass LSPs requires more intelligence on the server, and can substantially reduce the degree of distribution in case of a distributed setting. The use of the PCE based computation model, in this context, requires accounting for such dependency. The proposed solution is to regroup together all links whose protection placement must be coordinated into a new entity called Shared SRLG Dependency Link Group (SDLG). These links are said SRLG dependant. The result of such grouping is a set of disjoint groups, called Shared SRLG Dependency Link Groups, and noted SDLG. Vasseur, Le Roux et al. Expires December 2004 [Page 19] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 Then, in the context of the PCE based computation model, we extend the notion of facility to SDLGs. Each SDLG is treated, as a single element and is protected as a single entity (as a link or node), but with a modified aggregate bandwidth constraints, in order to take into account the assumption that only one SRLG fails and thus that not all bypass tunnels protecting a given SDLG are activated simultaneously. This is discussed in more detail below. 8.1. Notion of SRLG dependency, and Shared SRLG Dependency Link Group (SDLG) To take into account, in the facility based computation model, links that take part of multiple SRLGs, we define the notion of SRLG dependency: two links are said SRLG dependant, in the context of the facility based computation model, if their protection cannot be computed independently, or in other words if the computation of the NHOP bypass tunnels protecting these links must be done in a coordinated manner. It is clear that if two links are part of the same SRLG then they are SRLG dependant, but this is not necessary. Two SRLG diverse links maybe SRLG dependant, indeed in the above example, L1 and L3 are SRLG diverse but SRLG dependant. Note that this dependency relation is transitive. It means that if L1 and L2 are dependant and L2 and L3 are dependant then L1 and L3 are dependant. We define a Shared SRLG Dependency Link Group, noted SDLG, as a group of SRLG dependant links. An SDLG regroups all links that are SRLG dependant. From the transitivity property mentioned above, a link cannot belong to two SDLGs. Thus, it results that every link of a network, part of one ore more SRLGs, can be associated with a unique SDLG. The union of all the disjoint SDLGs is the set of links in the network. The number of SDLGs will depend on the repartition of SRLGs among network links. The number of SDLGs is always less than the number of SRLGs. At most (best case), nb SDLG = nb SRLG: this corresponds in fact to the particular case where all network links are part of 0 or one SRLG. At least (worst case) nb SDLG =1: it is the case where all SRLGs are linked, i.e. we cannot find two disjoint SRLGs. It is worth pointing out that a SDLG is no more than a union of linked SRLGs (ie a union of non disjoint SRLGs). An SDLG can be viewed as a union of SRLGs whose bandwidth protection computation must be done in a coordinated manner. Vasseur, Le Roux et al. Expires December 2004 [Page 20] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 Thus a SDLG is noted S1 U S2 ... U Sk. This significantly simplifies the manipulation of SDLGs by LSRs, and the algorithm to determine the set of SDLGs. The identification of SDLGs is required in a distributed computation. We propose to use as SDLG id, the lowest id of the union of SRLGs that compose the SDLG. 8.2. SDLG protection The key idea to support links that belong to multiple SRLGs, in the PCE based computation model, is to treat an SDLG as a single element, and protect it as a single entity (as links or node). The placement of the set of bypass tunnels protecting links from an SDLG is performed independently of any other element. The procedure is then relatively similar to the one for other elements (links or nodes). The computation of the set of tunnels protecting links of an SDLG, is performed in a coordinated manner, ignoring bandwidth of any bypass LSP protecting a distinct element (link, node or SDLG). The only distinction relies on the aggregate bandwidth constraint. Bypass tunnels computed for protection of an SDLG may protect different SRLGs. Thus, assuming than only one SRLG fails simultaneously, these bypass tunnels are not all activated simultaneously and it results that the aggregate bandwidth constraint on a link is lower than the cumulated bypass bandwidth. It is in fact the maximal bandwidth protecting an SRLG The PCE SHOULD take this specific aggregate bandwidth constraint into account when computing the placement of bypass tunnels corresponding to an SDLG to maximize the bandwidth sharing ratio. It is clear that the problem it has to solve is algorithmically more challenging than the simple problem of the placement of given bandwidth demands on a network of given topology. Here the problem it has to solve is how to find a feasible placement for a set of NON- ALL-SIMULTANEOUS flows of given demands, in a network of given topology. Both the centralized and distributed scenarios are supported. The centralized scenario requires no modification to what is defined in section 7.1, except the addition of the specific aggregate bandwidth constraint. By contrast, distributed computation requires a procedure specific to SDLGs that is specified in the section bellow. Vasseur, Le Roux et al. Expires December 2004 [Page 21] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 8.2.1. Distributed scenario for SDLGs protection. The same approach as defined in 6.2.3, is used to achieve a distributed SDLG protection. We propose that one of the end-nodes of the links forming the SDLG, be elected as PCS for whole SDLG. By default, the node with the lowest IGP id serves as PCS for the whole SDLG. PLR processing: - A PLR dynamically finds the SDLG its adjacent links belong to. - Then it determines for each SDLG, the corresponding PCE (ie the end-node with the lowest IGP id), and sends a Path computation request to these PCEs, indicating the SDLG id Note 1: In the particular case where all links are part of zero or one SRLG, a SDLG is reduced to a single SRLG, and the resulting distributed setting is then identical to what is proposed in 6.2.3. Thus SDLG protection supports networks were links belong to 0 or one SRLG. Note 2: In case all links are SRLG dependent, there is only one SDLG, and the result is a centralized computation (single PCE). Note 3: As soon as there is one link in the network that belongs to multiple SRLGs, the SDLG approach must be used. 8.3. Alternative solution An alternative solution to solve the problem of the computation of NHOP bypass tunnels protecting links part of multiple SRLGs could be to simply compute separate bypass LSP per SRLG for links belonging to multiple SRLGs. If the PLR could detect, upon the failure of a link, which of the SRLGs to which the link belongs actually failed, it could then use the appropriate bypass tunnel. In this case, each SRLG could be protected independently. However, this approach clearly requires that a PLR is capable of determining which SRLG actually fails when it observes a failure of a link belonging to multiple SRLGs. Unfortunately, no mechanism to identify which of the SRLGs actually failed currently exists. 9. Operations with DS-TE and multiple Class-Types This section assumes the reader is familiar with Diff-Serv-aware MPLS Traffic Engineering as specified in [DSTE-REQTS] and [DSTE-PROTO] and with its associated concepts such as Class-Types (CTs), Bandwidth Constraints (BCs) and the Russian Dolls bandwidth constraint model defined in [RDM]. Vasseur, Le Roux et al. Expires December 2004 [Page 22] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 The bandwidth protection approach described in this document supports DS-TE and operations with multiple Class-Types. It is worth mentioning that both the primary and backup bandwidth pools sizes have to be carefully determined by the network administrator as their values dictate the congestion level in case of failure, as discussed below. In the absence of failure, up to the max reservable bandwidth pool (i.e the primary bandwidth pool) of (primary) traffic will be forwarded onto a link. In case of failure, potentially up to "Primary bandwidth pool" + "backup bandwidth pool" of traffic will be active on a link. Various scenarios as to what the backup bandwidth should be reserved for, are discussed in the following sections. The determination of their values compared to the link speed is a critical factor. 9.1. Single backup pool Several bandwidth protection scenarios only require the use of a single backup pool. First, when a single Class-Type is used (i.e. network which do not use Diff-Serv or use Diff-Serv but only enforce a single bandwidth constraint to all the TE tunnels), bandwidth protection can be achieved via a single backup bandwidth pool. Second, when multiple Class-Types are used, a single backup pool can be used to provide bandwidth protection to LSPs from a single Class- Type CTc, which is the active CT with the highest index c, (in other words the active CT with the smallest Bandwidth Constraint), while LSPs from the other Class-Types do not get bandwidth protection. Here is an example of such scenario. Let's consider the following network where: - DS-TE and the Russian Dolls bandwidth constraint model are used - two Class-Types (CTs) are used: o CT1 is used for Voice Traffic o CT0 is used for Data traffic From a bandwidth protection perspective, let's assume that: - Voice traffic (i.e. CT1 LSPs) requires Bandwidth Protection during failure - Data traffic (i.e. CT0 LSPs) does not need Bandwidth Protection during failure. Let's further assume that the network administrator has elected to use the notion of backup pool and specify bandwidth requirements for Vasseur, Le Roux et al. Expires December 2004 [Page 23] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 bypass tunnels based on the full bandwidth pool of primary tunnels (i.e. BC1) as configured towards the protected facility (as opposed to the amount of bandwidth currently used by the primary LSPs requiring bandwidth protection). Then, for every link the network administrator will configure: - BC0, the Bandwidth Constraint on the aggregate across all primary LSPs (CT0+CT1) - BC1, the Bandwidth Constraint for primary CT1 LSPs - BCbu, the Bandwidth Constraint for the Backup CT1 LSPs The bandwidth requirement of each backup LSP is configured based on the value of BC1 configured towards the facility it protects. In other words, the backup LSPs are only sized to protect voice traffic transiting via the protected facility. Purely for illustration purposes, the diagram below represent these bandwidth constraints in a pictorial manner. I-------------------------------------------I ---------------I I--------------I I I I CT1 I I I I Primary I I I I--------------I I CT1 Backup I I CT1 + CT0 I I I-------------------------------------------I ---------------I I-----BC1------> I--------------------------------BC0------> I----BCbu-------> Note that while this scenario assumes Data traffic does not need Bandwidth protection during failure, Data traffic can be either not protected at all by Fast Reroute or be protected by Fast Reroute but without bandwidth protection during failure. In the former case, CT0 LSPs transporting Data traffic would not be rerouted into backup LSPs on failure. In the latter case, CT0 LSPs would be rerouted onto backup LSPs upon failure; the bypass tunnels could either be a different set of bypass tunnel from the bypass tunnels for voice, or could be the same bypass tunnels as for Voice assuming appropriate DiffServ marking and scheduling differentiation are configured properly, as discussed below. From a scheduling perspective, a possible approach is for Voice traffic to be treated as the exact same Ordered Aggregate (i.e. use the same EF PHB) whether it is transported on primary LSPs or on backup LSPs. In that case, on a given link, BC1 and BCbu must clearly be configured in such a way that the Voice QoS objectives are met when there is simultaneously, on that link, up to BC1 worth of traffic on primary CT1 LSPs and up to BCbu worth of Voice Traffic on backup LSPs. Vasseur, Le Roux et al. Expires December 2004 [Page 24] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 The size of the backup pool BCbu is configured on all links such that the CT1 LSP QoS objectives are met when there is simultaneously, on that link, up to BC1 worth of primary LSPs and up to BCbu worth of backup CT1 traffic. Notes - If the objective for CT1 traffic is only to protect CT1 bandwidth then the network administrator must just make sure that: BC1+BCbuLink Speed, CT0 traffic may experiment congestion during failure but CT1 traffic is still bandwidth-protected. Other scenarios can be addressed with a single bandwidth pool. This includes the case where all Class-Types need bandwidth protection but it is acceptable to relax delay guarantee to these classes during the failure and only offer bandwidth protection. Operations is very similar to the previous scenario described (e.g. size bypass tunnel based on BC0), the only difference is that QoS objectives other than bandwidth guarantee of other CTs than CT0 are not necessarily guaranteed to be preserved during failure. These CTs only get bandwidth assurances. 9.2. Multiple backup pools When DS-TE is used and multiple Class-Types are supported, the operations described above can be easily extended to multiple bandwidth pools in the case where backup LSPs are sized based on the actual amount of established LSPs: one backup pool can be used to separately constrain the bandwidth used by backup LSPs of each Class- Type. In that case, each CT can be given bandwidth protection during failure with guarantee that each CT will also meet all its respective QoS objectives during the failure and without any bandwidth wastage. Here is an example of such scenario. Let's consider the following network where: - DS-TE and the Russian Dolls bandwidth constraint model are used - two Class-Types (CTs) are used: o CT1 is used for Voice Traffic o CT0 is used for Data traffic From a bandwidth protection perspective, let's assume that: - Voice traffic (i.e. CT1 LSPs) needs Bandwidth Protection during failure - Data traffic (i.e. CT0 LSPs) also needs Bandwidth Protection during failure. Let's further assume that the network administrator has elected to Vasseur, Le Roux et al. Expires December 2004 [Page 25] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 specify bandwidth requirements for bypass tunnels based on the actual amount of established primary LSPs requiring bandwidth protection (as opposed to the full bandwidth pool of primary tunnels as configured towards the protected facility). Then, for every link the network administrator will configure: - BC0, the Bandwidth Constraint on the aggregate across all primary LSPs (CT0+CT1) - BC1, the Bandwidth Constraint for primary CT1 LSPs - BCbu0, the Bandwidth Constraint on the aggregate across all backup LSPs (CT0+CT1) - BCbu1, the Bandwidth Constraint on the CT1 backup LSPs The bandwidth requirement of each CT0 backup LSP is configured based on the actual amount of established CT0 primary LSPs it protects. The bandwidth requirement of each CT1 backup LSP is configured based on the actual amount of established CT1 primary LSPs it protects. Purely for illustration purposes, the diagram below represents these bandwidth constraints in a pictorial manner. I--------------------------------------I--------------------I I--------------I I----------I I I CT1 I I CT1 I I I Primary I I Backup I I I--------------I I----------I I I CT1 + CT0 Primary I CT1+CT0 Backup I I--------------------------------------I--------------------I I-----BC1------> I--BCbu1--> I----------------------------BC0------>I-------BCbu0-------> The size of the backup pool BCbu0 is configured on all links such that the CT0 LSP QoS objectives are met when there is simultaneously, on that link, up to BC0 worth of CT0 primary LSPs and up to BCbu0 worth of backup CT0 traffic. The size of the backup pool BCbu1 is configured on all links such that the CT1 LSP QoS objectives are met when there is simultaneously, on that link, up to BC1 worth of CT1 primary LSPs and up to BCbu1 worth of backup CT1 traffic. In the case where backup LSPs are sized based on the amount of reservable bandwidth, it is also possible to extend operations to multiple bandwidth pools in the same way, but this may result in bandwidth wastage. This is because BC1 will be effectively reserved both from BC1bu and from BC0bu (with the RDM model). Operations with multiple backup pools, as well as operations with the Maximum Allocation Model [MAM]), will be discussed in more details in subsequent versions of this draft. Vasseur, Le Roux et al. Expires December 2004 [Page 26] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 10. Interaction with scheduling The bandwidth protection approach described in this document does not require any enhancement or modification to MPLS scheduling mechanisms beyond those defined in [MPLS-DIFF]. In particular, scheduling can remain entirely unaware of Fast Reroute and bandwidth protection; in particular this approach does not require that scheduling behave differently depending on whether a packet is transported on a primary LSP or a backup LSP, nor does it require per-LSP scheduling. This approach simply requires that the existing MPLS scheduling mechanisms (e.g. Diff-Serv PHBs) are configured in a manner which is compatible with the goal of bandwidth protection, because while the bandwidth protection allocates bandwidth appropriately in the control plane, it is scheduling which is responsible for the actual enforcement in the data path of the corresponding service rates to packets in a way which will achieve the targeted bandwidth protection. The details of which configuration is appropriate depends on multiple parameters such as the details of the Diff-Serv policy, the bandwidth protection policy and the number of DS-TE Class-Types supported. Thus, it is outside the scope of this draft. For illustration purpose, the uniform Diff-Serv tunneling mode defined in section 2.6 of [MPLS-DIFF] may be used on the bypass tunnels, for bandwidth protected LSPs. In particular, when a packet is steered into a bypass tunnel by the PLR (i.e. when the bypass tunnel label entry is pushed onto the packet) the EXP field of the packet is copied into the EXP field of the bypass tunnel label entry. In the particular case where the PLR could not establish a bypass tunnel with the full requested amount of bandwidth (due to some lack of bandwidth in the backup pool) and instead established a bypass tunnel with a smaller bandwidth, when rerouting LSPs onto this bypass tunnel, the PLR may ensure that the amount of rerouted primary LSPs complies with the actual bandwidth of the bypass tunnel. This can done using the same bypass tunnel (or a separate bypass tunnel) with the pipe DiffServ tunneling mode for the non bandwidth protected primary rerouted TE LSPs (this both includes the set of TE LSPs not requiring bandwidth protection and the set of TE LSP that have required bandwidth protection but for which there was not enough backup bandwidth on the bypass tunnel to accommodate their request). Otherwise, this would simply violate bandwidth protection (for traffic on this bypass tunnel as well as for all traffic on any LSP using the same PHBs) because more traffic than reserved for would end up in the bypass tunnel. Vasseur, Le Roux et al. Expires December 2004 [Page 27] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 11. PLR-PCE Signaling: Functional description A LSR-PCE signaling protocol is required for PLR-PCE communication. The PLR will have to send a request to the PCE, for the computation of a bypass path. Such request will be characterized by the specification of several parameters that are discussed bellow. 11.1. Element to protect The PLR specifies the element to protect: Link, Node, SRLG or SDLG. Typically, a link protection request will result in a set of NHOP bypass tunnels as a node protection request will result in a set of NNHOP bypass tunnels. 11.2. Bandwidth to protect There are two different approaches for the bandwidth-to-protect parameter: - The bypass tunnel bandwidth may be based on the amount of reservable bandwidth pool on a particular network resource, - The bypass tunnel bandwidth may be based on the sum of bandwidths actually reserved by established TE LSPs requiring bandwidth protection on a particular resource. 11.3. Affinities The requesting node may also specify affinities constraint. Affinities for the bypass tunnel may be configured on the PLR by the network administrator or derived from the FAST-REROUTE object of the protected TE LSP, if used. In this former case, this would require some rules to derive the affinities of the bypass tunnel from the affinities of the protected TE LSPs making use of this bypass tunnel. 11.4. Maximum number of bypass tunnels It may happen that no single bypass tunnel can fulfil the constraints requirements. In such a situation, a set of bypass tunnels could be computed such that the sum of the bandwidths of every element in the set is at least equal to the required bandwidth. It may be desirable to bound the number of elements in this set by specifying a maximum number of bypass tunnels originating at a PLR and protecting an element. 11.5. Minimum bandwidth on any element of a set of bypass tunnels When a solution can be found with a set of bypass tunnels it may also Vasseur, Le Roux et al. Expires December 2004 [Page 28] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 be desirable to provide some constraint on the minimal bandwidth value for any bypass tunnel in the set. As an example, if a 100M bypass tunnel is required, a set of 1000 tunnels each having 100K is likely to be unacceptable. Also, it is worth reminding that a single protected TE LSP will make use of a single bypass tunnel at a given time. 11.6. Class Type to protect Specifies the Class-Type(s) to protect. See section 9 on operations with DS-TE. 11.7. Set of already in place bypass tunnels In certain circumstances (stateless PCE), it may also be useful for the PLR to provide to the PCE the set of already in place bypass tunnels with their corresponding constraints for the PCE to try to minimize the incremental changes of the existing bypass tunnels due to the placement of new bypass tunnels. 12. Bypass tunnel - Make before break In case of bypass tunnel path change, the new bypass tunnel may be set up using make before break. This may or not be possible depending on the change in the set of bypass tunnels. 13. Stateless versus Statefull PCE There are basically two options for the PCE: - can be statefull: the PCE registers the various bypass tunnels computation requests and results. It will also monitor the network states (bypass tunnels in place, ...) - can be stateless: the PCE does not maintain any state. This approach is the recommended approach for the distributed model. 14. Packing algorithm Once the set of bypass tunnels is in place and their respective bandwidth, the PLR should, for each protected TE LSP successfully signaled, select a corresponding bypass tunnel. As per defined in [FAST-REROUTE], the bandwidth protection requirement for the protected LSP can be specified using the FAST-REROUTE object or by setting the ''Bandwidth protection desired'' bit in the SESSION- ATTRIBUTE of the Path message. Based on the signaled backup bandwidth requirement for the protected LSP, the PLR should appropriately select the bypass tunnel to use for the protected TE LSP, making sure the requested backup bandwidth requirement is met. Vasseur, Le Roux et al. Expires December 2004 [Page 29] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 15. Security Considerations The practice described in this draft does not raise specific security issues beyond those of existing TE. 16. Acknowledgements The authors would like to thank Carol Iturralde, Rog Goguen, Vishal Sharma, Shahram Davari and Renaud Moignard for their useful comments in a former version of this draft. 17. Security Considerations No new security issues are raised in this document. 18. Intellectual Property Statement The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79. Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr. The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf- ipr@ietf.org.. 18.1. IPR Disclosure Acknowledgement By submitting this Internet-Draft, I certify that any applicable patent or other IPR claims of which I am aware have been disclosed, or will be disclosed, and any of which I become aware will be disclosed, in accordance with RFC 3668. Vasseur, Le Roux et al. Expires December 2004 [Page 30] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 19. References Normative references [RFC] Bradner, S., "Key words for use in RFCs to indicate requirements levels", RFC 2119, March 1997. [RFC] Bradner, S., "Key words for use in RFCs to indicate requirements levels", RFC 2119, March 1997. [RFC3667] Bradner, S., "IETF Rights in Contributions", BCP 78, RFC 3667, February 2004. [RFC3668] Bradner, S., Ed., "Intellectual Property Rights in IETF Technology", BCP 79, RFC 3668, February 2004. [TE-REQ] Awduche et al, Requirements for Traffic Engineering over MPLS, RFC2702, September 1999. Informative Reference [FAST-REROUTE] Pan, P. et al., "Fast Reroute Techniques in RSVP-TE", Internet Draft, draft-ietf-mpls-rsvp-lsp-fastreroute-06.txt, work in progress [BP-PLACEMENT] Le Roux, J.L., Calvignac, G., "A method for an Optimized Online Placement of MPLS Bypass Tunnels", draft-leroux- mpls-bypass-placement-00.txt, February 2002. [KINI] Kini et al, "Shared Backup Label Switched Path Restoration", draft-kini-restoration-shared-backup-01.txt, May 2001. [MPLS-DIFF] Le Faucheur et al, "Multi-Protocol Label Switching (MPLS) Support of Differentiated Services", RFC 3270, May 2002. [RDM] Le Faucheur et al., "Russian Dolls Bandwidth Constraints Model for Diff-Serv-aware MPLS Traffic Engineering", draft-ietf-tewg-diff- te-russian-06.txt, work in progress. [MAM] Le Faucheur, F., Lai, W., "Maximum Allocation Bandwidth Constraints Model for Diff-Serv-aware MPLS Traffic Engineering" , draft-ietf-tewg-diff-te-mam-03.txt, work in progress. [LINKNODE-FAILURE] Vasseur, Charny, "Distinguish a link from a node failure using RSVP Hellos extensions", draft-vasseur-mpls-linknode- failure-00.txt, November 2002. [RFC3469] Sharma V., et al, "Framework for Multi-Protocol Label Switching (MPLS)-based Recovery", Feb 2003. Vasseur, Le Roux et al. Expires December 2004 [Page 31] Internet Draft draft-leroux-pce-backup-comp-frwk-00 July 2004 20. Authors' Address: Jean Philippe Vasseur Cisco Systems, Inc. 300 Beaver Brook Road Boxborough , MA - 01719 USA Email: jpv@cisco.com Anna Charny Cisco Systems, Inc. 300 Beaver Brook Road Boxborough , MA - 01719 USA Email: acharny@cisco.com Francois Le Faucheur Cisco Systems, Inc. Village d'Entreprise Green Side - Batiment T3 400, Avenue de Roumanille 06410 Biot-Sophia Antipolis FRANCE Email: flefauch@cisco.com Javier Achirica Telefonica Data Espana Beatriz de Bobadilla, 14 28040 Madrid SPAIN Email: javier.achirica@telefonica-data.com Jean-Louis Le Roux France Telecom 2, avenue Pierre-Marzin 22307 Lannion Cedex FRANCE E-mail: jeanlouis.leroux@francetelecom.com Full Copyright Statement "Copyright (C) The Internet Society (2004). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights." "This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE." Vasseur, Le Roux et al. Expires December 2004 [Page 32]