Internet Draft IETF Draft Changcheng Huang Multi-Protocol Label Switching Vishal Sharma Expires: September 2000 Srinivas Makam Ken Owens Tellabs March 2000 A Path Protection/Restoration Mechanism for MPLS Networks <draft-chang-mpls-path-protection-00.txt. Status of this memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet- Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Abstract To deliver reliable service, multi-protocol label switching (MPLS)0 requires a set of procedures to provide protection of the traffic carried on the label switched paths (LSPs). This imposes certain requirements on the path recovery process and procedures 0, and requires - signaling support for: the configuration of working and protection paths, the communication of fault information, and appropriate switchover action. This document specifies a mechanism for path protection switching and restoration in MPLS. Table of Contents 1.0 Introduction 2.0 Core Path Protection Components 2.1 Reverse Notification Tree (RNT) 2.2 Protection Domain 2.2.1 Relationship between protection domains with different RNTs 2.2.2 Relationship between protection domains with the same RNT 2.3 Multiple Faults 2.4 Timers and Thresholds 3.0 Configuration 3.1 Establishing a Recovery/Protection Path 3.2 Creating the RNT 3.3 Engineering a Protection Domain 3.4 Configuring Timers 4.0 Fault Detection 4.1 Unidirectional Link Fault 4.1.1 Downlink Fault 4.1.2 Uplink Fault 4.2 Bi-directional Link Fault or -Node Fault 5.0 Fault Notification 6.0 Switch Over 7.0 Switch Back 8.0 Security considerations 9.0 Acknowledgement 10.0 Intellectual Property Consideration 11.0 Authors' Addresses 12.0 References 1.0 Introduction With the migration of real-time and high-priority traffic to IP networks, and with the need for IP networks to increasingly carry mission-critical business data, network survivability has become critical for future IP networks. Current routing algorithms, despite being robust and survivable, can take a substantial amount of time, on the order of several seconds to minutes, to recover from a failure causing serious disruption of service in the interim. This is unacceptable for many applications that require a highly reliable service, and has motivated network providers to give serious consideration to the issue of network survivability. Path-oriented technologies, such as MPLS, can be used to support advanced survivability requirements and enhance the reliability of IP networks. Different from legacy IP networks, MPLS networks pre- establish label switched paths (LSPs), with packets with the same label following the same path. This potentially allows MPLS to pre- establish protection LSPs for working LSPs, and achieve better protection switching times thanlegacy IP networks. This contribution describes an MPLS path recovery mechanism that can facilitate fast protection switching. The mechanism supports both 1:1 and bypass tunneling, and contains timers to enable it to inter-work with protection mechanisms at other layers. Some of the key features of this protection mechanism are: A liveness message to detect faults. Our assumption is that faults fall into different classes, and that different faults may be detected and corrected by different layers. Some faults (for example, the loss of signal or transmitter faults) may be detected and corrected by lower layer mechanisms (such as SONET), while others (for example, failure of the reverse link) may be detected (but may not be corrected) by lower layers and may be communicated to the MPLS layer. Still other faults (such as node failures or faults on the reverse link) may not be detected by lower layers, and will have to be detected and corrected at the MPLS layer. Therefore, we adopt the liveness message as a complementary fault detection mechanism. - Special tree structure to distribute fault and/or recovery information. Existing published proposals û for MPLS recovery have not addressed the issue of fault notification in detail. Specifically, none of these proposals has discussed how to - perform fault notification for the label merging case-. In this draft, we propose a new fault notification ûstructure, called the reverse notification tree (RNT), which - makes fault notification more efficient and ûscalable (we provide details of the RNT in subsequent sections). - Ability to permit recovery mechanisms at different layers to coexist. In the evolving IP network infrastructure, recovery will increasingly be possible at different layers, and the interworking of recovery mechanisms at different layers will be needed to ensure smooth network operation in the presence of faults. For example, optical layer or SONET layer recovery mechanisms could be used to recover optical paths or SONET channels. However, these mechanisms may initially be limited to ring topologies and may not provide the right level of granularity at which recovery might be desired, making MPLS layer protection desirable. While MPLS-layer protection can be faster than IP layer rerouting it may be more costly to use (it may need to reserve extra bandwidth for example). In certain cases - it may become too complicated for MPLS protection to be effective (for example, when there are multiple faults, or faults on both the working and protection paths), making it necessary for IP layer rerouting û to take over -. In this draft, we assume that MPLS layer protection is - is used first, and, if necessary, recovery may be handed - to IP layer rerouting following that. - Lightweight notification mechanism. Reliable transport mechanisms, such as TCP, are typically state- oriented and therefore difficult to scale. It is also very difficult to support point-to-multipoint communications based on reliable transport mechanisms. In our scheme, therefore, we use a stateless notification mechanism to achieve scalability. - Minimize delays of a recovery cycle. In the mechanism proposed in this draft we attempt to minimize the ûduration of the recovery cycle. To minimize the notification delay we use a stateless transport mechanism together with high priority for the control traffic. We also use a simple label merging approach to handle the traffic on the working and protection paths, thereby eliminating the need for synchronization (or handshaking) between the LSRs at the two ends of a recovery path. 2.0 Core MPLS Path Protection Components This document assumes the terminology given in 0,0 and 0, and introduces some additional terms. - Path Switch LSR (PSL) An LSR that is the transmitter of both the working path traffic and its corresponding recovery path traffic. The PSL is responsible for switching of the traffic between the working path and the recovery path. The PSL is the origin of the recovery path, but may or may not be the origin of the working path (that is the working path may be transiting the PSL). - Path Merge LSR (PML) An LSR that receives both working path traffic and its corresponding recovery path traffic, and either merges their traffic into a single outgoing path, or, if it is itself the destination, passes the traffic on to the higher layer protocols. The PML is the destination of the recovery path, but may or may not be the destination of the working path (that is, the working path may be transiting through the PML) - Intermediate LSR An LSR on a working or recovery path that is neither a PSL nor a PML for that path. - Working Path A working path is denoted by the sequence of LSRs through which it passes. For example, in Fig. 1, the working path that starts at LSR 1 and terminates at LSR 7 is denoted by (1-2-3-4-6-7). - Recovery Path A recovery path is also denoted by the sequence of LSRs through which it passes. Again, in Fig. 1, the recovery path that starts at LSR 1 and terminates at LSR 7 is denoted by (1-5-7). 2.1 Reverse Notification Tree (RNT) Since LSPs are unidirectional entities and recovery requires the notification of faults to the LSR responsible for switchover to the recovery path, a mechanism must be provided for the fault indication and the fault recovery notification to travel from the point of occurrence of the fault back to the PSL(s). The situation is complicated with label merging, because in this case multiple working paths converge to form a multipoint-to-point tree, with the PSLs as the leaves and the PML as the root. In this case, therefore, the fault indication and recovery notification should be able to travel along a reverse path of the working path to all the PSLs affected by the fault. Such a path is provided by the reverse notification tree (RNT), which is a point-to-multipoint tree rooted at the PML that is an exact mirror image of the converged multipoint-to-point working paths, along which the FIS and the FRS travel (see Fig. 1). There are several advantages to having an RNT: - The RNT can be established in association with the working path, simply by making each LSR along a working path remember its upstream neighbor (or the collection of upstream neighbors whose working paths converge at the LSR and exit as one. Thus, no multicast routing is required. We elaborate more on the RNT in Section 3. - Only one RNT is required for all the working paths that merge to form the multipoint-to-point forward path. The RNT is rooted at the PML and terminated at the PSLs. All intermediate LSRs on the converged working paths share the same RNT. Therefore, the RNT enables a reduction in the signaling overhead associated with recovery. Unlike schemes that treat each LSP independently, and require signaling between a PSL and the PML for each LSP individually, the RNT allows for only one (or a small number of) signaling messages on the shared segments of the LSPs. - The RNT can be implemented either at Layer 3 or at Layer 2. In either case, the delay along the RNT needs to be carefully controlled. This may be ensured by giving the highest priority to the fault and recovery notification packets, which travel along the RNT. 2.2 Protection Domain The protection domain - is defined by the set of LSRs over which the working path and its corresponding recovery path are routed. Thus, a protection domain is bounded by the LSRs that provide the switching and merging functions for MPLS protection-, namely, the PSL and the PML, respectively. The PSL and the PML are identified during the setting up of an LSP, either via an offline algorithm or an algorithm that runs at the head-end of an LSP to decide the specific nodes that the LSP must pass through. (Note that segments of the LSP between the PSL and the PML may be loosely-explicitly routed, as long as the PSL and PML are known). Recovery should ideally be performed between the source and destination (end-to-end), but in some cases segment recovery may be desired (for example, when certain segments are more unreliable than others)or may be the only option (due to the topology of the network, see Fig. 1). For example, in Fig. 1, the working path 9-3-4-6-7, can only have protection on the segment 9-3-4-6-7. Note that when multiple LSPs merge into a single LSP, the working paths corresponding to these LSPs also converge. As explained in Section 2.4, an RNT is needed in this case for propagating the failure and recovery notification back to the concerned PSL(s). We can therefore have a situation where different protection domains share a common RNT. - A protection domain is denoted by a specifying the working path and the recovery path. For example, in Fig. 1, the protection domain bounded by LSR 1 and LSR 7, is denoted by (1-2-3-4- 6-7, 1-5-7). Figure 1: Illustration of MPLS protection configuration. 2.2.1 Relationship between protection domains with different RNTs When protection domains have different RNTs, two cases may arise, depending on whether or not any portions of the two domains overlap, that is, have a node or link in common. If the protection domains do not overlap, the protection domains are independent (note that by virtue of the RNTs in the two domains being different, neither the working paths nor the RNTs in the two domains can overlap). In other words, failures in one domain do not interact with failures in the other domain. For example, the protection domain defined by (9-3-4-6- 7, 9-10-7) is completely independent of the domain defined by (11-13- 5-15, 11-13-14-15). As a result, as long as faults occur in independent domains, the network shown in Fig. 1 can tolerate multiple -faults (for example, simultaneous failures on the working path in each domain). If protection domains with different RNTs overlap, it is still the case that failures on the working paths of the two domains do not affect one another. However, failures on the protection path of one may affect the working path of the other and visa versa. For example, the protection domain defined by (1-2-3-4-6-7, 1-5-7) is not independent of the domain defined by (11-13-5-15, 11-13-14-15 ) since LSR 5 lies on the protection path of the former domain and on the working path of the latter domain. 2.2.2 Relationship between protection domains with the same RNT When protection domains have the same RNT, different failures along the working paths may affect both paths differently. As shown in Fig. 1, for example, working paths 1-2-3-4-5-7 and 9-3-4-6-7 share the same RNT. As a result, for a failure on some segment of the working path, both domains will be affected, resulting in a protection switch in both (for example, the segment 3-4-6-7 in Fig. 1). Likewise, for failures on other segments of the working path, only one domain may be affected (for example, failure on segment 2-3 affects only the first working path 1-2-3-4-6-7, where as failure on the segment 9-3 affects only the second working path 9-3-4-6-7). 2.3 Multiple Faults We note that transferring the working traffic to the recovery path is enough to take care of multiple faults on the working path. However, if multiple faults happen such that there is at least one failure on both the working and recovery paths, MPLS layer recovery may no longer suffice. In this case, the PSL will either have to allow for Layer 3 rerouting or inform the administrator via an alarm, thus enabling the manual reconfiguration of a different working and backup path. Note that for the PSL to be able to generate an alarm, it must have a mechanism for detecting multiple faults on the recovery path, such as a RNT for the recovery path (to allow for the fault notification on the recovery path to be propagated to the PSL). 2.4 Timers and Thresholds For its proper operation, the protection mechanism described in this contribution uses the following timers and thresholds: Timer or Symbol Function Threshold Inter FIS t1 Interval at which successive FIS packet timer packets are transmitted by a LSR to its upstream neighbor. Max. FIS t2 Max. time for which FIS packets are duration timer transmitted by an LSR to its upstream peer. Inter FRS T1Æ Interval at which successive FRS packet timer packets are sent by a LSR to its upstream neighbor. Max. FRS t2Æ Max. time for which the FRS packets duration timer are sent by an LSR to its upstream neighbor. Protection t3 Time interval between receipt of a switching protection switch trigger and the dampening timer initiation of the protection switch, thereby allowing traffic on the working path (downstream of the fault) to clear out. Restoration t3Æ Time interval between the initiation dampening timer of the restoration switch and the actual flow of data on the working path, thereby allowing traffic on the recovery path to clear out. Liveness msg. t4 Interval at which successive liveness sender timer messages are sent by an LSR to peer LSRs that have a working path (and RNT) through this LSR. Liveness msg. t4' A timer set to count down the receiver interval at the end of which a timeout timer liveness message should be received. Hold-off Timer T2 Interval between the detection of a 0 failure at an LSR, and the generation of the first FIS message, to allow time for lower layer protection to take effect. Wait-to-Restore T8 Interval between the detection of a Timer 0 recovery/failure at an LSR, and the generation of the first FRIS message, to allow time for the stability of restoration. Lost liveness K No. of liveness messages that can be message lost before an LSR will declare a threshold fault and generate the first FIS. Table 1. Timers and Thresholds 3.0 Configuration In the following sections, we describe the operation of the path protection mechanism, and explain the various steps involved with reference to Fig. 1. Protection configuration consists of two aspects: establishing the protection path and creating the reverse notification tree. 3.1 Establishing a Recovery/Protection Path The establishment of the recovery path requires the identification of the working path, and hence the protection domain. In most cases, the working path and its corresponding recovery path would be specified during LSP setup, either via a path selection algorithm (running at a centralized location or at an ingress LSR) or via administrative configuration. Observe that the specification of the path, does not, strictly speaking, require the entire path to be explicitly specified. Rather, it requires only that the PSL and PML be specified, with the segments between them being be loosely routed, if required. In other words, the path would be established between the two nodes at the boundaries of the protection domain (namely, the PSL and the PML) via explicit (or source) routing using LDP 0, 0/RSVP 0, 0 signaling (alternatively, via constraint-based routing (with the requirement that the path pass through the PSL and the PML), or using manual configuration). The signaling would be used to specify both PMTPs and working paths, where the working paths could span either an entire LSP or a segment of a LSP. Ingress Ingress Egress Egress Egress Egress Label of Interface Label of Interface Label of Interface RNT of RNT RNT of RNT RNT of RNT N43 I34 N32 I23 N39 I93 Table 2. An example inverse cross-connect table for LSR 3 using MPLS (Layer 2) RNT Egress Egress Next Hop Egress Next Hop Egress Label of Interface IP Address Interface IP Address Interface Working of Working of RNT of RNT of RNT of RNT Path Path L34 I34 I2 I23 I9 I93 Table 3. An example inverse cross-connect table for LSR 3 using hop- by-hop (Layer 3) RNT The roles of the various core protection/recovery components are: PSL: The PSL initiates the working LSP and the recovery LSP. It is also responsible for storing information about which LSPs (or portions thereof) have protection enabled, and for maintaining a binding between outgoing labels -specifying the working path and the protection/recovery path. The latter enables the switchover to the recovery path upon the receipt of a protection switch trigger. The PSL also maintains the timers t3, t3Æ, t4, t4Æ, T2, T8, and the threshold K. PML: The PML participates in the setting up of a recovery path as a merging LSR. Therefore, it learns during signaling (or configuration) about which working and protection paths are merged to the same outgoing LSP. The PML also maintains timers t1, t1',t2, t2Æ, t4, t4', T2, T8, and the threshold K. Intermediate LSR: An intermediate LSR participates in the setup of the recovery path, either as a normal LSR or as a merging LSR. It also maintains timers t1, t1', t2, t2Æ, t4, t4Æ, T2, T8, and the threshold K. 3.2 Creating the RNT The RNT is used for propagating the FIS and the FRS, and can be created by a simple extension to the LSP setup process. During the establishment of the working path, the signaling message carries with it the identity (address) of the upstream node that sent it (-for example, via the path attribute in RSVP). Each LSR along the path simply remembers the identity of its immediately prior upstream neighbor on each incoming link. Through neighbor discovery mechanism of the routing protocol, each LSR findsthe interface connecting it to an upstream LSR. (It is assumed in this draft that there is a bi- directional connection between two neighboring LSRs, such as a bi- directional SONET link a bi-directional lower layer network link (e.g., an ATM VP), or a pair of bi-directional tunnels over IP an subnetwork. The node then creates an ôinverseö cross-connect table that for each protected outgoing LSP maintains a list of the incoming LSPs that merge into that outgoing LSP, together with the identity of the upstream node and incoming interface that each incoming LSP comes - through. Upon receiving an FIS, an LSR extracts the labels contained in it (which are the labels of the protected LSPs that use the outgoing link that the FIS was received on) and checks whether the current LSR is the PSL for that LSP. If it is it terminates the FIS. Otherwise, it consults its inverse cross-connect table to determine the identity of the upstream nodes that the protected LSPs come from, and creates and transmits an FIS to each of them. Therefore, based on whether the RNT is implemented at Layer 3 or Layer 2, two cases arise: If the RNT is implemented by a point-to-multipoint LSP, then the working path can be bound to the ingress label and interface of the RNT LSP at a LSR. The ingress label and interface then can be used as an index in the "inverse" cross-connect table to find the egress labels and interfaces of the RNT LSP as shown in Table 2. Upon receiving an FIS, an LSR extracts the labels and checks whether it is the PSL for that LSP. If it is, it terminates the FIS. Otherwise, it consults its inverse cross-connect table to determine the outgoing labels and interfaces, inserts them into the FIS and forwards it to the appropriate upstream node(s). If the RNT is implemented by a hop-by-hop Layer 3 mechanism, using, for example, UDP packets(with a specific port number to identify notification message type), then the egress label and interface of the working path can be used as an index into the inverse cross- connect table to obtain the IP addresses of the previous hop(s) and the associated outgoing interface(s), as shown in Table 3. On each hop, the FIS carried in the UDP packet will carry the label and interface of the working path for that hop. Thus, if the receiving node is not a PSL, the label and interface in the FIS can be extracted to access the inverse cross-connect table, and the label and interface used by the working LSP on the hop(s) to the upstream node(s) can be inserted into FIS packet(s). The FIS packet(s) are then transmitted to the appropriate upstream node(s). The roles of the various core protection/recovery components are: PSL: The PSL must be able to correlate the RNT with the working and recovery paths. To this end, it maintains a table with a list of working LSPs protected by an RNT, and the identity of the recovery LSPs that each working path is to be switched to in the event of a failure on the working path. It need not maintain an inverse cross- connect table (for those LSPs and working paths for which it is the PSL). PML: The PML is the root of the RNT, and has to associate each of its upstream nodes with a working path and RNT. It need not maintain an inverse cross-connect table (for those LSPs and working paths for which it is a PML). Intermediate LSR: An intermediate LSR has to only remember all of its upstream neighbors and associate them with the appropriate working paths and RNTs, and has to maintain an ôinverseö cross-connect table. 3.3 Engineering a Protection Domain For 1:1 protection, the bandwidth reserved for a protection/recovery path should be the same as the bandwidth reserved for its corresponding working path. This guarantees the same bandwidth for the protected traffic after protection switching. If the LSRs on the protection path support excess mode 0, the bandwidth reserved on the protection path for protecting high priority traffic can be used by other lower priority traffic streams. That is, low priority traffic that is destined for the same node as the working traffic can be sent on the protection path, and is transmitted onward by the PML after merging with the working traffic. Also, if delay, jitter or other QoS parameters are to be satisfied, the protection path in 1:1 protection should be chosen such that these requirements are satisfied. Since the volume of signaling traffic (e.g., FIS/FRS messages, or liveness messages) is small, in general bandwidth need not be reserved for the signaling traffic provided that there exist other mechanisms that can ensure that the delay requirements of signaling messages are met (by using, for example, the highest priority for signaling messages). For bypass tunneling protection, multiple working LSPs may share the same protection bandwidth by tunneling protection LSPs over a common path. This requires that e the paths of these working LSPs be disjoint, except at the PSL and PML, so that they can be assumed to not all fail at the same time. In this case, the bandwidth reserved will be the maximum of all individual paths. Otherwise, a bypass tunnel could be created to carry all the backup paths, with the bandwidth reserved for the tunnel being the maximum bandwidth required over all failure scenarios of the working LSPs. 3.4 Configuring Timers The purpose of timers t1/t1' is to control the tradeoff between notification delay of the FIS/FRS and the resources consumed when sending the FIS/FRS. If t1/t1' is large, it may take a relatively long time for the initiation node to send the second the FIS/FRS if the first FIS/FRS message is lost, thereby increasing notification delay. On the other hand, if t1/t1' is small, the repetitive sending of FIS/FRS messages may waste bandwidth and processing power because the first message may already have reached the PSL(s). It is assumed that after t2/t2' it is not necessary to do protection at MPLS layer, either because it is no longer useful or because by that time an upper layer protection mechanism will have been triggered. The purpose of timers t3/t3' is to minimize the misordering of packets at a PML following a protection (restoration) switch from the working (backup) to the backup (working) path. This is because packets buffered on the working (backup) path, downstream of the fault, may continue to arrive at the PML even as working traffic begins to arrive on the protection (working) path. Therefore, forcing the PSL to hold off the protection (or restoration) switching action, gives the buffers on the working (protection) path time to clear before data on the protection (working) path begins to arrive. The timers t4/t4' are used to control the frequency of liveness messages sent between neighboring LSRs where t4 control how often the liveness message should be sent out from the sender side and t4' is the timeout timer on the receiver side. While frequent exchanges of liveness messages can unnecessarily consume network resources, too few exchanges may delay the discovery of faults. To accommodate delay jitter, t4' may be set at a slightly different value from t4. The timers T2/T8 are used to allow the lower layer protection to take effect before initiating MPLS layer recovery mechanisms (for example, an automatic protection switching between fibers that comprise a link between two LSRs). Following the detection of a fault/fault recovery, an LSR waits for T2/T8 time units before issuing the first FIS/FRS packet, respectively. This allows for the lower layer protection to take effect and for the LSR to learn this through one of several ways: via an indication from a lower layer, or by the resumption of the reception of a liveness message, or by the lack ofLF, LD, PF or PD conditions. The threshold K helps to minimize false alarms due to the occasional loss of a liveness message, which may occur, for example, due to a temporary impairment in a link or a peer LSR or due to a buffer overflow. 4.0 Fault Detection Each LSR must be able to detect certain types of faults, such as PF, PD, LF, and LD 0 and propagate an FIS message towards the PSL. Here we consider unidirectional linkfaults , bi-directional (or complete) linkfaults , and nodefaults. 4.1 Unidirectional Link Fault A uni-directional link fault implies that only one direction of a bi- directional link has experienced a fault. 4.1.1 Downlink Fault A fault on a link in the downstream direction will be detected by the node downstream of the faulty link, either via the PF or PD condition being detected at the MPLS layer, or via LF or LD signals being propagated to the MPLS layer by the lower layer or via the absence of liveness messages. The downstream node will then periodically transmit FIS messages to its upstream neighbor (via the uplink), which will propagate these further upstream (using its inverse cross- connect table) until they eventually reach the appropriate PSLs, which will perform the protection switch. Therefore, in Fig. 1, if link L34 has a fault,LSR 4 will detect the fault via one of the means described above, and start transmitting an FIS packet once every t1 time units back to LSR 3 over link L43. The traffic in the queues of LSR 4 will continue to be serviced. LSR 3 in turn will propagate the FIS over the RNT back to LSR 2 and LSR 9. The actual protection switch will be performed by LSRs 9 and 1, t3 time units after the receipt of the first FIS. LSR 4 will stop transmitting FIS messages t2 time units after the transmission of the first FIS message. 4.1.2 Uplink Fault A fault on a link in the upstream direction will be detected by a node upstream of the faulty link, either via a LF or LD being detected at the lower layer and propagated to the MPLS layer (if there was traffic on this reverse link), or via the PD or PF condition being detected at the MPLS layer, or via absence of liveness messages. The upstream node will then periodically send out FIS messages to the node upstream of it, which in turn will propagate these further until eventually the PSL(s) learns of the failure and performs the protection switch. Therefore, in Fig. 1, if link L43 experiences a fault, LSR 3 will detect the fault, and transmit an FIS to nodes 2 and 9. Node 2, in turn, will transmit an FIS to node 1, and nodes 1 and 9 will perform the actual protection switch 4.2 Bi-directional link fault or Node Fault When both directions of the link have a fault (as in the case of a fiber cut), nodes at both ends of the link will detect the fault either due to the LF or PF signal or due to the absence of liveness messages. Both will transmit FIS messages to their upstream nodes. However, it is only the node upstream of the failed link whose FIS messages will propagate further upstream, eventually reaching the appropriate PSLs, which will perform the protection switch to the recovery path. The case of a node fault is similar, with the node upstream of the failed node detecting the failure (due to loss of liveness messages, for example) and propagating that information via the FIS message. For example in Fig. 1, when both directions of the link between nodes 3 and 4 experience a fault (or when node 4has a fault), LSR 3 will detect this failure via the non-reception of the liveness message, and transmit FIS messages to nodes 2 and 9 as before. When nodes 1 and 9 receive the FIS message they will perform the protection switch after waiting for an interval of t3 time units. The roles of the various core protection components in failure detection are the same. The PSL, PML, and intermediate LSR must all be able to detect PF and PD conditions and/or be able to interpret and respond to the LF and LD indications received from the lower layers. 5.0 Fault Notification The rapid notification of a fault is effected by the propagation of the FIS message along the RNT. Due to the timers built into the FIS/FRS propagation mechanism, the transportation of FIS/FRS messages does not require a reliable mechanism like TCP. Any LSR may generate an FIS, but a PSL is the only LSR that may terminate it. For instance, in Fig. 1 if link L23 fails, LSR 3 will detect it and transmit a FIS to LSR 2 (after waiting for time T2), its upstream neighbor along link L23. The FIS will contain the incoming labels (at node 3) of those LSPs on link L23 that have protection enabled. Upon receiving the FIS message, LSR 2 will consult its inverse-cross connect table and generate an FIS message for LSR 1, which on receiving the first FIS packet will wait for time t3 before performing a protection switch. The node which initiates the FIS will continue to send FIS messages at an interval of t1 until timer t2 expires. After t2 expires it is assumed that either upper layer protection will be triggered or enough number of FIS messages will have been sent to reach the desired reliability in conveying fault information to the PSL(s). The roles of the various core protection switching components are: PSL: The PSL does not generate a FIS message, but must be able to detect FIS packets. PML: The PML must be able to generate the FIS packets in response to detecting failure, and should transmit them over the RNT. The PML begins FIS transmission after continuously detecting a fault for T2 time units, and does so every t1 time units for a maximum of t2 time units. Intermediate LSR: An intermediate LSR must be able to generate/forward FIS packets, either as a result of continuously detecting a fault for T2 time units or in response to a received FIS packet. It must transmit these to all its affected upstream neighbors as per its inverse cross-connect table. Again, it does so every t1 time units for a maximum of t2 time units. 6.0 Switch Over The switch over is the actual switching of the working traffic from the working path to the recovery path. This is performed by a PSL, t3 time units after the reception of the first FIS packet. For example, in Fig. 1, consider protection domain (1-2-3-4-6-7, 1-5- 7). When link L34 fails, the PSL LSR 1 on learning of the failure will perform a protection switch of the protected traffic from the working path 1-2-3-4-6-7 to the backup path 1-5-7. Notice that LSR 7 acts as a protection merge LSR, merging traffic from the working and backup paths. Since buffered packets from LSR 4 may continue to arrive at LSR 7 even after the protection switch (the dampening timer t43at the PSL tends to mitigate this), a short-term misordering of packets may happen at LSR 7, until the buffers on the working path drain out. The role of the core protection components is as follows: PSL: Performs the protection switch upon receipt of the FIS message, but after waiting for time t3 following the first FIS message. PML: The PML automatically merges protection traffic with working traffic. For a short period of time this may cause misordering of packets, since packets buffered at LSRs downstream of the fault may continue to arrive at the PML along the working path. Intermediate LSR: The intermediate LSR has no special action. 7.0 Switch Back Switch back or restoration is the transfer of working traffic from the recovery path to the working path, once the working path is repaired. This may be because the recovery path may be a limited recovery path 0, or because the working path is deemed to be preferred 0in some respect. Restoration may be automatic or it may be performed by manual intervention (or not performed at all). In the revertive mode, restoration is performed upon the receipt of the FRS message, while in the non-revertive mode it may be performed by operator intervention. The role of the core protection components is similar here to what it is for protection switching. The PML does not need to do anything, unless it was the node that detected the failure, in which case it transmits a FRS upstream T8 time units after continuously detecting recover signal from lower layer or after detecting liveness messages from its peers. The intermediate LSR generates the FRS message if it was the node that detected the recovery or generates a FRS to relay the restoration status received from a downstream node. The PSL performs the restoration switch t3Æ seconds after receiving the first FIS message. 8.0 Security Considerations The MPLS protection that is specified herein does not raise any security issues that are not already present in the MPLS architecture. 9.0 Acknowledgement We would like to thank Mr. Ben Mack-Crane from Tellabs. 10.0 Intellectual Property Consideration In accordance with the intellectual property rights procedures of the IETF standards process, to the extent that Tellabs has patents, pending applications and/or other intellectual property rights that are essential to implementation of any subject matter submitted by Tellabs that is included in a standard, Tellabs is prepared to grant, on the basis of reciprocity (grantback), a license on such subject matter under terms and conditions that are reasonable and non- discriminatory. 11.0 Authors' Addresses Changcheng Huang Tellabs Operations, Inc. 4951 Indiana Avenue Lisle, IL 60532 Email: Changcheng.Huang@tellabs.com Ph: 630-512-7954 Vishal Sharma Tellabs Research Center One Kendall Square Bldg. 100, Suite 121 Cambridge, MA 02139 Email: Vishal.Sharma@trc.tellabs.com Ph: 617-577-8760 Srinivas Makam Tellabs Operations, Inc. 4951 Indiana Avenue Lisle, IL 60532 Email: Srinivas.Makam@tellabs.com Ph: 630-512-7217 Ken Owens Tellabs 1106 Fourth Street St. Louis, MO 63126 Email: Ken.Owens@tellabs.com Ph: 314-825-7009 12.0 References [1] Rosen, E., Viswanathan, A., and Callon, R., "Multiprotocol Label Switching Architecture", Work in Progress, Internet Draft, August 1999. [2] Callon, R., Doolan, P., Feldman, N., Fredette, A., Swallow, G., Viswanathan, A., "A Framework for Multiprotocol Label Switching", Work in Progress, Internet Draft <draft-ietf-mpls-framework-05.txt>, September 1999. [3] Awduche, D., Malcolm, J., Agogbua, J., O'Dell, M., McManus,J., "Requirements for Traffic Engineering Over MPLS", RFC 2702, September 1999. [4] Andersson, L., Doolan, P., Feldman, N., Fredette, A., Thomas, B., "LDP Specification", Work in Progress, Internet Draft , September 1999. [5] Jamoussi, B. "Constraint-Based LSP Setup using LDP", Work in Progress, Internet Draft <draft-ietf-mpls-cr-ldp-03.txt>, September 1999. [6] Makam, V., Sharma, V., Huang, C., Owens, K., Mack-Crane, B., et al, ôA Framework for MPLS-based Recovery,ö Work in Progress, Internet Draft <draft-makam-mpls-recovery-frmwrk-00.txt>, February 2000. [7] Braden, R., Zhang, L., Berson, S., Herzog, S., "Resource ReSerVation Protocol (RSVP) -- Version 1 Functional Specification", RFC 2205, September 1997. [8] Awduche, D. et al "Extensions to RSVP for LSP Tunnels", Work in Progress, Internet Draft <draft-ietf-mpls-rsvp-lsp-tunnel-04.txt, September 1999.