RFC 6372
MPLS Transport Profile (MPLS-TP) Survivability Framework
Pages: 56
Informational
Informational
Part 2 of 3 – Pages 10 to 37
4. Functional Architecture
This section presents an overview of the elements relating to the functional architecture for survivability within an MPLS-TP network. The components are presented separately to demonstrate the way in which they may be combined to provide the different grades of recovery needed to meet the requirements set out in the previous section.4.1. Elements of Control
Recovery is achieved by implementing specific actions. These actions aim to repair network resources or redirect traffic along paths that avoid failures in the network. They may be triggered automatically by the MPLS-TP network nodes upon detection of a network defect, or they may be triggered by an operator. Automated actions may be enhanced by in-band (i.e., data-plane-based) OAM mechanisms, or by in-band or out-of-band control-plane signaling.
4.1.1. Operator Control
The survivability behavior of the network as a whole, and the reaction of each transport path when a fault is reported, may be controlled by the operator. This control can be split into two sets of functions: policies and actions performed when the transport path is set up, and commands used to control or force recovery actions for established transport paths. The operator may establish network-wide or local policies that determine the actions that will be taken when various defects are reported that affect different transport paths. Also, when a service request is made that causes the establishment of one or more transport paths in the network, the operator (or requesting application) may define a particular grade of service, and this will be mapped to specific survivability actions taken before and during transport path setup, after the discovery of a failure of network resources, and upon recovery of those resources. It should be noted that it is unusual to present a user or customer with options directly related to recovery actions. Instead, the user/customer enters into an SLA with the network provider, and the network operator maps the terms of the SLA (for example, for guaranteed delivery, availability, or reliability) to recovery schemes within the network. The operator can also issue commands to control recovery actions and events. For example, the operator may perform the following actions: o Enable or disable the survivability function. o Invoke the simulation of a network fault. o Force a switchover from a working path to a recovery path or vice versa. Forced switchover may be performed for network optimization purposes with minimal service interruption, such as when modifying protected or unprotected services, when replacing MPLS-TP network nodes, etc. In some circumstances, a fault may be reported to the operator, and the operator may then select and initiate the appropriate recovery action. A description of the different operator commands is found in Section 4.12 of [RFC4427].
4.1.2. Defect-Triggered Actions
Survivability actions may be directly triggered by network defects. This means that the device that detects the defect (for example, notification of an issue reported from equipment in a lower layer, failure to receive an OAM Continuity message, or receipt of an OAM message reporting a failure condition) may immediately perform a survivability action. The action is directly triggered by events in the data plane. Note, however, that coordination of recovery actions between the edges of the recovery domain may require message exchanges for some recovery functions or for performing a bidirectional recovery action.4.1.3. OAM Signaling
OAM signaling refers to data-plane OAM message exchange. Such messages may be used to detect and localize faults or to indicate a degradation in the operation of the network. However, in this context these messages are used to control or trigger survivability actions. The mechanisms to achieve this are discussed in [RFC6371]. OAM signaling may also be used to coordinate recovery actions within the protection domain.4.1.4. Control-Plane Signaling
Control-plane signaling is responsible for setup, maintenance, and teardown of transport paths that do not fall under management-plane control. The control plane may also be used to coordinate the detection, localization, and reaction to network defects pertaining to peer relationships (neighbor-to-neighbor or end-to-end). Thus, control-plane signaling may initiate and coordinate survivability actions. The control plane can also be used to distribute topology and information relating to resource availability. In this way, the "graceful shutdown" [RFC5817] of resources may be affected by withdrawing them; this can be used to invoke a survivability action in a similar way to that used when reporting or discovering a fault, as described in the previous sections. The use of a control plane for MPLS-TP is discussed in [RFC6373].
4.2. Recovery Scope
This section describes the elements of recovery. These are the quantitative aspects of recovery, that is, the parts of the network for which recovery can be provided. Note that the terminology in this section is consistent with [RFC4427]. Where the terms differ from those in [RFC5654], mapping is provided.4.2.1. Span Recovery
A span is a single hop between neighboring MPLS-TP nodes in the same network layer. A span is sometimes referred to as a link, and this may cause some confusion between the concept of a data link and a traffic engineering (TE) link. LSPs traverse TE links between neighboring MPLS-TP nodes in the MPLS-TP network layer. However, a TE link may be provided by any of the following: o A single data link. o A series of data links in a lower layer, established as an LSP and presented to the upper layer as a single TE link. o A set of parallel data links in the same layer, presented either as a bundle of TE links, or as a collection of data links that together provide a data-link-layer protection scheme. Thus, span recovery may be provided by any of the following: o Selecting a different TE link from a bundle. o Moving the TE link so that it is supported by a different data link between the same pair of neighbors. o Rerouting the LSP in the lower layer. Moving the protected LSP to another TE link between the same pair of neighbors is a form of segment recovery and not a form of span recovery. Segment Recovery is described in Section 4.2.2.4.2.2. Segment Recovery
An LSP segment comprises one or more continuous hops on the path of the LSP. [RFC5654] defines two terms. A "segment" is a single hop along the path of an LSP, while a "concatenated segment" is more than one hop along the path of an LSP. In the context of this document, a segment covers both of these concepts.
A PW segment refers to a Single-Segment PW (SS-PW) or to a single segment of a Multi-Segment PW (MS-PW) that is set up between two PE devices that may be Terminating PEs (T-PEs) or Switching PEs (S-PEs) so that the full set of possibilities is T-PE to S-PE, S-PE to S-PE, S-PE to T-PE, or T-PE to T-PE (for the SS-PW case). As indicated in Section 1, the recovery of PWs and PW segments is beyond the scope of this document; however, see Section 7. Segment recovery involves redirecting or copying traffic at the source end of a segment onto an alternate path leading to the other end of the segment. According to the required grade of recovery (described in Section 4.3), traffic may be either redirected to a pre-established segment, through rerouting the protected segment, or tunneled to the far end of the protected segment through a "bypass" LSP. For details on recovery mechanisms, see Section 4.4. Note that protecting a transport path against node failure requires the use of segment recovery or end-to-end recovery, while a link failure can be protected using span, segment, or end-to-end recovery.4.2.3. End-to-End Recovery
End-to-end recovery is a special case of segment recovery where the protected segment comprises the entire transport path. End-to-end recovery may be provided as link-diverse or node-diverse recovery where the recovery path shares no links or no nodes with the working path. Note that node-diverse paths are necessarily link-diverse and that full, end-to-end node-diversity is required to guarantee recovery. Two observations need to be made about end-to-end recovery. - Firstly, there may be circumstances where node-diverse end-to-end paths do not guarantee recovery. The ingress and egress nodes will themselves be single points of failure. Additionally, there may be shared risks of failure (for example, geographic collocation, shared resources, etc.) between diverse nodes as described in Section 4.9.2. - Secondly, it is possible to use end-to-end recovery techniques even when there is not full diversity and the working and protection paths share links or nodes.
4.3. Grades of Recovery
This section describes the qualitative grades of survivability that can be provided. In the event of a network failure, the grade of recovery offered directly affects the service grade provided to the end-user. This will be observed as the amount of data lost when a network fault occurs, and the length of time required to recover connectivity. In general, there is a correlation between the recovery service grade (i.e., the speed of recovery and reduction of data loss) and the amount of resources used in the network; better service grades require the pre-allocation of resources to the recovery paths, and those resources cannot be used for other purposes if high-quality recovery is required. An operator will consider how providing different grades of recovery may require that network resources be provisioned and allocated for exclusive use of the recovery paths such that the resources cannot be used to support other customer services. Sections 6 and 7 of [RFC4427] provide a full breakdown of the protection and recovery schemes. This section summarizes the qualitative grades available. Note that, in the context of recovery, a useful discussion of the term "resource" and its interpretation in both the IETF and ITU-T contexts may be found in Section 3.2 of [RFC4397]. The selection of the recovery grade and schemes to satisfy the service grades for an LSP using available network resources is subject to network and local policy and may be pre-designated through network planning or may be dynamically determined by the network.4.3.1. Dedicated Protection
In dedicated protection, the resources for the recovery entity are pre-assigned for the sole use of the protected transport path. This will clearly be the case in 1+1 protection, and may also be the case in 1:1 protection where extra traffic (see Section 4.3.3) is not supported. Note that when using protection tunnels (see Section 4.4.3), resources may also be dedicated to the protection of a specific transport path. In some cases (1:1 protection), the entire bypass tunnel may be dedicated to providing recovery for a specific transport path, while in other cases (such as facility backup), a subset of the resources associated with the bypass tunnel may be pre- assigned for the recovery of a specific service.
However, as described in Section 4.4.3, the bypass tunnel method can also be used for shared protection (Section 4.3.2), either to carry extra traffic (Section 4.3.3) or to achieve best-effort recovery without the need for resource reservation.4.3.2. Shared Protection
In shared protection, the resources for the recovery entities of several services are shared. These may be shared as 1:n or m:n and are shared on individual links. Link-by-link resource sharing may be managed and operated along LSP segments, on PW segments, or on end- to-end transport paths (LSP or PW). Note that there is no requirement for m:n recovery in the list of MPLS-TP requirements documented in [RFC5654]. Shared protection can be applied in different topologies (mesh, ring, etc.) and can utilize different protection mechanisms (linear, ring, etc.). End-to-end shared protection shares resources between a number of paths that have common end points. Thus, a number of paths (n paths) are all protected by one or more protection paths (m paths, where m may equal 1). When there have been m failures, there are no more available protection paths, and the n paths are no longer protected. Thus, in 1:n protection, one fault can be protected against before all the n paths are unprotected. The fact that the paths have become unprotected needs to be conveyed to the path end points since they may need to report the change in service grade or may need to take further action to increase their protection. In end-to-end shared protection, this communication is simple since the end points are common. In shared mesh protection (see Section 4.7.6), the paths that share the protection resources do not necessarily have the same end points. This provides a more flexible resource-sharing scheme, but the network planning and the coordination of protection state after a recovery action are more complex. Where a bypass tunnel is used (Section 4.4.3), the tunnel might not have sufficient resources to simultaneously protect all of the paths for which it offers protection; in the event that all paths were affected by network defects and failures at the same time, not all of them would be recovered. Policy would dictate how this situation should be handled: some paths might be protected, while others would simply fail; the traffic for some paths would be guaranteed, while traffic on other paths would be treated as best-effort with the risk of dropped packets. Alternatively, it is possible that protection would not be attempted according to local policy at the nodes that perform the recovery actions.
Shared protection is a trade-off between assigning network resources
to protection (which is not required most of the time) and risking
unrecoverable services in the event that multiple network defects or
failures occur. Rapid recovery can be achieved with dedicated
protection, but it is delayed by message exchanges in the management,
control, or data planes for shared protection. This means that there
is also a trade-off between rapid recovery and resource sharing. In
some cases, shared protection might not meet the speed required for
protection, but it may still be faster than restoration.
These trade-offs may be somewhat mitigated by the following:
o Adjusting the value of n in 1:n protection.
o Using m:n protection for a value of m > 1.
o Establishing new protection paths as each available protection
path is put into use.
In an MPLS-TP network, the degree to which a resource is shared
between LSPs is a policy issue. This policy may be applied to the
resource or to the LSPs, and may be pre-configured, configured per
LSP and installed during LSP establishment, or may be dynamically
configured.
4.3.3. Extra Traffic
Section 2.5.1.1 of [RFC5654] says: "Support for extra traffic (as
defined in [RFC4427]) is not required in MPLS-TP and MAY be omitted
from the MPLS-TP specifications". This document observes that extra
traffic facilities may therefore be provided as part of the MPLS-TP
survivability toolkit depending upon the development of suitable
solution specifications. The remainder of this section explains the
concepts of extra traffic without prejudging the decision to specify
or not specify such solutions.
Network resources allocated for protection represent idle capacity
during the time that recovery is not actually required, and can be
utilized by carrying other traffic, referred to as "extra traffic".
Note that extra traffic does not need to start or terminate at the
ends of the entity (e.g., LSP) that it uses.
When a network resource carrying extra traffic is required for the
recovery of protected traffic from the failed working path, the extra
traffic is disrupted. This disruption make take one of two forms:
- In "hard preemption", the extra traffic is excluded from the
protection resource. The disruption of the extra traffic is total,
and the service supported by the extra traffic must be dropped, or
some form of rerouting or restoration must be applied to the extra
traffic LSP in order to recover the service.
Hard preemption is achieved by "setting a switch" on the path of
the extra traffic such that it no longer flows. This situation may
be detected by OAM and reported as a fault, or may be proactively
reported through OAM or control-plane signaling.
- In "soft preemption", the extra traffic is not explicitly excluded
from the protection resource, but is given lower priority than the
protected traffic. In a packet network (such as MPLS-TP), this can
result in oversubscription of the protection resource with the
result that the extra traffic receives "best-effort" delivery.
Depending on the volume of protection and extra traffic, and the
level of oversubscription, the extra traffic may be slightly or
heavily impacted.
The event of soft preemption may be detected by OAM and reported as
a degradation of traffic delivery or as a fault. It may also be
proactively reported through OAM or control-plane signaling.
Note that both hard and soft preemption may utilize additional
message exchanges in the management, control, or data planes. These
messages do not necessarily mean that recovery is delayed, but may
increase the complexity of the protection system. Thus, the benefits
of carrying extra traffic must be weighed against the disadvantages
of delayed recovery, additional network overhead, and the impact on
the services that support the extra traffic according to the details
of the solutions selected.
Note that extra traffic is not protected by definition, but may be
restored.
Extra traffic is not supported on dedicated protection resources,
which, by definition, are used for 1+1 protection (Section 4.3.1),
but it can be supported in other protection schemes, including shared
protection (Section 4.3.2) and tunnel protection (Section 4.4.3).
Best-effort traffic should not be confused with extra traffic. For
best-effort traffic, the network does not guarantee data delivery,
and the user does not receive guaranteed quality of service (e.g., in
terms of jitter, packet loss, delay, etc.). Best-effort traffic
depends on the current traffic load. However, for extra traffic,
quality can only be guaranteed until resources are required for
recovery. At this point, the extra traffic may be completely
displaced, may be treated as best effort, or may itself be recovered (for example, by restoration techniques).4.3.4. Restoration
This section refers to LSP restoration. Restoration for PWs is beyond the scope of this document (but see Section 7). Restoration represents the most effective use of network resources, since no resources are reserved for recovery. However, restoration requires the computation of a new path and the activation of a new LSP (through the management or control plane). It may be more time- consuming to perform these steps than to implement recovery using protection techniques. Furthermore, there is no guarantee that restoration will be able to recover the service. It may be that all suitable network resources are already in use for other LSPs, so that no new path can be found. This problem can be partially mitigated by using LSP setup priorities, so that recovery LSPs can preempt existing LSPs with lower priorities. Additionally, when a network defect occurs, multiple LSPs may be disrupted by the same event. These LSPs may have been established by different Network Management Stations (NMSes) or they may have been signaled by different head-end MPLS-TP nodes, meaning that multiple points in the network will try to compute and establish recovery LSPs at the same time. This can lead to a lack of resources within the network and cause recovery failures; some recovery actions will need to be retried, resulting in even slower recovery times for some services. Both hard and soft LSP restoration may be supported. For hard LSP restoration, the resources of the working LSP are released before the recovery LSP is fully established (i.e., break-before-make). For soft LSP restoration, the resources of the working LSP are released after an alternate LSP is fully established (i.e., make-before- break). Note that in the case of reversion (Section 4.3.5), the resources associated with the working LSP are not released. The restoration resources may be pre-calculated and even pre-signaled before the restoration action starts, but not pre-allocated. This is known as pre-planned LSP restoration. The complete establishment/activation of the restoration LSP occurs only when the restoration action starts. Pre-planning may occur periodically and provides the most accurate information about the available resources in the network.
4.3.5. Reversion
After a service has been recovered and traffic is flowing along the recovery LSP, the defective network resource may be replaced. Traffic can be redirected back onto the original working LSP (known as "reversion"), or it can be left where it is on the recovery LSP ("non-revertive" behavior). It should be possible to specify the reversion behavior of each service; this might even be configured for each recovery instance. In non-revertive mode, an additional operational option is possible where protection roles are switched, so that the recovery LSP becomes the working LSP, while the previous working path (or the resources used by the previous working path) are used for recovery in the event of an additional fault. In revertive mode, it is important to prevent excessive swapping between the working and recovery paths in the case of an intermittent defect. This can be addressed by using a reversion delay timer (the Wait-To-Restore timer), which controls the length of time to wait before reversion following the repair of a fault on the original working path. It should be possible for an operator to configure this timer per LSP, and a default value should be defined.4.4. Mechanisms for Protection
This section provides general descriptions (MPLS-TP non-specific) of the mechanisms that can be used for protection purposes. As indicated above, while the functional architecture applies to both LSPs and PWs, the mechanism for recovery described in this document refers to LSPs and LSP segments only. Recovery mechanisms for pseudowires and pseudowire segments are for further study and will be described in a separate document (see also Section 7).4.4.1. Link-Level Protection
Link-level protection refers to two paradigms: (1) where protection is provided in a lower network layer and (2) where protection is provided by the MPLS-TP link layer. Note that link-level protection mechanisms do not protect the nodes at each end of the entity (e.g., a link or span) that is protected. End-to-end or segment protection should be used in conjunction with link-level protection to protect against a failure of the edge nodes.
Link-level protection offers the following grades of protection:
o Full protection where a dedicated protection entity (e.g., a link
or span) is pre-established to protect a working entity. When the
working entity fails, the protected traffic is switched to the
protecting entity. In this scenario, all LSPs carried over the
working entity are recovered (in one protection operation) when
there is a failure condition. This is referred to in [RFC4427] as
"bulk recovery".
o Partial protection where only a subset of the LSPs or traffic
carried over a selected entity is recovered when there is a
failure condition. The decision as to which LSPs will be
recovered and which will not depends on local policy.
When there is no failure on the working entity, the protection entity
may transport extra traffic that may be preempted when protection
switching occurs.
If link-level protection is available, it may be desirable to allow
this to be attempted before attempting other recovery mechanisms for
the transport paths affected by the fault because link-level
protection may be faster and more conservative of network resources.
This can be achieved both by limiting the propagation of fault
condition notifications and by delaying the other recovery actions.
This consideration of other protection can be compared with the
discussion of recovery domains (Section 4.5) and recovery in multi-
layer networks (Section 4.9).
A protection mechanism may be provided at the MPLS-TP link layer
(which connects two MPLS-TP nodes). Such a mechanism can make use of
the procedures defined in [RFC5586] to set up in-band communication
channels at the MPLS-TP Section level, to use these channels to
monitor the health of the MPLS-TP link, and to coordinate the
protection states between the ends of the MPLS-TP link.
4.4.2. Alternate Paths and Segments
The use of alternate paths and segments refers to the paradigm
whereby protection is performed in the network layer in which the
protected LSP is located; this applies either to the entire end-to-
end LSP or to a segment of the LSP. In this case, hierarchical LSPs
are not used (compare with Section 4.4.3).
Different grades of protection may be provided:
o Dedicated protection where a dedicated entity (e.g., LSP or LSP
segment) is (fully) pre-established to protect a working entity
(e.g., LSP or LSP segment). When a failure condition occurs on
the working entity, traffic is switched onto the protection
entity. Dedicated protection may be performed using 1:1 or 1+1
linear protection schemes. When the failure condition is
eliminated, the traffic may revert to the working entity. This is
subject to local configuration.
o Shared protection where one or more protection entities is pre-
established to protect against a failure of one or more working
entities (1:n or m:n).
When the fault condition on the working entity is eliminated, the
traffic should revert back to the working entity in order to allow
other related working entities to be protected by the shared
protection resource.
4.4.3. Protection Tunnels
A protection tunnel is pre-provisioned in order to protect against a
failure condition along a sequence of spans in the network. This may
be achieved using LSP heirarchy. We call such a sequence a network
segment. A failure of a network segment may affect one or more LSPs
that transit the network segment.
When a failure condition occurs in the network segment (detected
either by OAM on the network segment, or by OAM on a concatenated
segment of one of the LSPs transiting the network segment), one or
more of the protected LSPs are switched over at the ingress point of
the network segment and are transmitted over the protection tunnel.
This is implemented through label stacking. Label mapping may be an
option as well.
Different grades of protection may be provided:
o Dedicated protection where the protection tunnel reserves
sufficient resources to provide protection for all protected LSPs
without causing service degradation.
o Partial protection where the protection tunnel has enough
resources to protect some of the protected LSPs, but not all of
them simultaneously. Policy dictates how this situation should be
handled: it is possible that some LSPs would be protected, while
others would simply fail; it is possible that traffic would be
guaranteed for some LSPs, while for other LSPs it would be treated
as best effort with the risk of packets being dropped.
Alternatively, it is possible that protection would not be
attempted.
4.5. Recovery Domains
Protection and restoration are performed in the context of a recovery domain. A recovery domain is defined between two or more recovery reference end points that are located at the edges of the recovery domain and that border on the element on which recovery can be provided (as described in Section 4.2). This element can be an end- to-end path, a segment, or a span. An end-to-end path can be observed as a special segment case where the ingress and egress Label Edge Routers (LERs) serve as the recovery reference end points. In this simple case of a point-to-point (P2P) protected entity, two end points reside at the boundary of the protection domain. An LSP can enter through one reference end point and exit the recovery domain through another reference end point. In the case of unidirectional point-to-multipoint (P2MP), three or more end points reside at the boundary of the protection domain. One of the end points is referred to as the source/root, while the others are referred to as sinks/leaves. An LSP can enter the recovery domain through the root point and exit the recovery domain through the leaf points. The recovery mechanism should restore traffic that was interrupted by a facility (link or node) fault within the recovery domain. Note that a single link may be part of several recovery domains. If two recovery domains have common links, one recovery domain must be contained within the other. This can be referred to as nested recovery domains. The boundaries of recovery domains may coincide, but recovery domains must not overlap. Note that the edges of a recovery domain are not protected, and unless the whole domain is contained within another recovery domain, the edges form a single point of failure. A recovery group is defined within a recovery domain and consists of a working (primary) entity and one or more recovery (backup) entities that reside between the end points of the recovery domain. To guarantee protection in all situations, a dedicated recovery entity should be pre-provisioned using disjoint resources in the recovery domain, in order to protect against a failure of a working entity. Of course, mechanisms to detect faults and to trigger protection switching are also needed. The method used to monitor the health of the recovery element is beyond the scope of this document. The end points that are
responsible for the recovery action must receive information on its condition. The condition of the recovery element may be 'OK', 'failed', or 'degraded'. When the recovery operation is to be triggered by OAM mechanisms, an OAM Maintenance Entity Group must be defined for each of the working and protection entities. The recovery entities and functions in a recovery domain can be configured using a management plane or a control plane. A management plane may be used to configure the recovery domain by setting the reference points, the working and recovery entities, and the recovery type (e.g., 1:1 bidirectional linear protection, ring protection, etc.). Additional parameters associated with the recovery process may also be configured. For more details, see Section 6.1. When a control plane is used, the ingress LERs may communicate with the recovery reference points that request that protection or restoration be configured across a recovery domain. For details, see Section 6.5. Cases of multiple interconnections between distinct recovery domains create a hierarchical arrangement of recovery domains, since a single top-level recovery domain is created from the concatenation of two recovery domains with multiple interconnections. In this case, recovery actions may be taken both in the individual, lower-level recovery domains to protect any LSP segment that crosses the domain, and within the higher-level recovery domain to protect the longer LSP segment that traverses the higher-level domain. The MPLS-TP recovery mechanism can be arranged to ensure coordination between domains. In interconnected rings, for example, it may be preferable to allow the upstream ring to perform recovery before the downstream ring, in order to ensure that recovery takes place in the ring in which the defect occurred. Coordination of recovery actions is particularly important in nested domains and is discussed further in Section 4.9.4.6. Protection in Different Topologies
As described in the requirements listed in Section 3 and detailed in [RFC5654], the selected recovery techniques may be optimized for different network topologies if the optimized mechanisms perform significantly better than the generic mechanisms in the same topology. These mechanisms are required (R91 of [RFC5654]) to interoperate with the mechanisms defined for arbitrary topologies, in order to allow
end-to-end protection and to ensure that consistent protection techniques are used across the entire network. In this context, 'interoperate' means that the use of one technique must not inhibit the use of another technique in an adjacent part of the network for use on the same end-to-end transport path, and must not prohibit the use of end-to-end protection mechanisms. The next sections (4.7 and 4.8) describe two different topologies and explain how recovery may be markedly different in those different scenarios. They also develop the concept of a recovery domain and show how end-to-end survivability may be achieved through a concatenation of recovery domains, each providing some grade of recovery in part of the network.4.7. Mesh Networks
A mesh network is any network where there is arbitrary interconnectivity between nodes in the network. Mesh networks are usually contrasted with more specific topologies such as hub-and- spoke or ring (see Section 4.8), although such networks are actually examples of mesh networks. This section is limited to the discussion of protection techniques in the context of mesh networks. That is, it does not include optimizations for specific topologies. Linear protection is a protection mechanism that provides rapid and simple protection switching. In a mesh network, linear protection provides a very suitable protection mechanism because it can operate between any pair of points within the network. It can protect against a defect in a node, a span, a transport path segment, or an end-to-end transport path. Linear protection gives a clear indication of the protection status. Linear protection operates in the context of a protection domain. A protection domain is a special type of recovery domain (see Section 4.5) associated with the protection function. A protection domain is composed of the following architectural elements: o A set of end points that reside at the boundary of the protection domain. In the simple case of 1:n or 1+1 P2P protection, two end points reside at the boundary of the protection domain. In each transmission direction, one of the end points is referred to as the source, and the other is referred to as the sink. For unidirectional P2MP protection, three or more end points reside at the boundary of the protection domain. One of the end points is referred to as the source/root, while the others are referred to as sinks/leaves.
o A Protection Group consists of one or more working (primary) paths
and one or more protection (backup) paths that run between the end
points belonging to the protection domain. To guarantee
protection in all scenarios, a dedicated protection path should be
pre-provisioned to protect against a defect of a working path
(i.e., 1:1 or 1+1 protection schemes). In addition, the working
and the protection paths should be disjoint; i.e., the physical
routes of the working and the protection paths should be
physically diverse in every respect.
Note that if the resources of the protection path are less than those
of the working path, the protection path may not have sufficient
resources to protect the traffic of the working path.
As mentioned in Section 4.3.2, the resources of the protection path
may be shared as 1:n. In this scenario, the protection path will not
have sufficient resources to protect all the working paths at a
specific time.
For bidirectional P2P paths, both unidirectional and bidirectional
protection switching are supported. If a defect occurs when
bidirectional protection switching is defined, the protection actions
are performed in both directions (even if the defect is
unidirectional). The protection state is required to operate with a
level of coordination between the end points of the protection
domain.
In unidirectional protection switching, the protection actions are
only performed in the affected direction.
Revertive and non-revertive operations are provided as options for
the network operator.
Linear protection supports the protection schemes described in the
following sub-sections.
4.7.1. 1:n Linear Protection
In the 1:1 scheme, a protection path is allocated to protect against
a defect, failure, or a degradation in a working path. As described
above, to guarantee protection, the protection entity should support
the full capacity and bandwidth, although it may be configured (for
example, because of limited network resource availability) to offer a
degraded service when compared with the working entity.
Figure 1 presents 1:1 protection architecture. In normal conditions,
data traffic is transmitted over the working entity, while the
protection entity functions in the idle state. (OAM may run on the
protection entity to verify its state.) Normal conditions are defined when there is no defect, failure, or degradation on the working entity, and no administrative configuration or request causes traffic to flow over the protection entity. |-----------------Protection Domain---------------| ============================== /**********Working path***********\ +--------+ ============================== +--------+ | Node /| |\ Node | | A {< | | >} B | | | | | +--------+ ============================== +--------+ Protection path ============================== Figure 1: 1:1 Protection Architecture If there is a defect on the working entity or a specific administrative request, traffic is switched to the protection entity. Note that when operating with non-revertive behavior (see Section 4.3.5), after the conditions causing the switchover have been cleared, the traffic continues to flow on the protection path, but the working and protection roles are not switched. In each transmission direction, the protection domain source bridges traffic onto the appropriate entity, while the sink selects traffic from the appropriate entity. The source and the sink need to coordinate the protection states to ensure that bridging and selection are performed to and from the same entity. For this reason, a signaling coordination protocol (either a data-plane in- band signaling protocol or a control-plane-based signaling protocol) is required. In bidirectional protection switching, both ends of the protection domain are switched to the protection entity (even when the fault is unidirectional). This requires a protocol to coordinate the protection state between the two end points of the protection domain. When there is no defect, the bandwidth resources of the idle entity may be used for traffic with lower priority. When protection switching is performed, the traffic with lower priority may be preempted by the protected traffic through tearing down the LSP with lower priority, reporting a fault on the LSP with lower priority, or by treating the traffic with lower priority as best effort and discarding it when there is congestion.
In the general case of 1:n linear protection, one protection entity is allocated to protect n working entities. The protection entity might not have sufficient resources to protect all the working entities that may be affected by fault conditions at a specific time. In this case, in order to guaranteed protection, the protection entity should support enough capacity and bandwidth to protect any of the n working entities. When defects or failures occur along multiple working entities, the entity to be protected should be prioritized. The protection states between the edges of the protection domain should be fully coordinated to ensure consistent behavior. As explained in Section 4.3.5, revertive behavior is recommended when 1:n is supported.4.7.2. 1+1 Linear Protection
In the 1+1 protection scheme, a fully dedicated protection entity is allocated. As depicted in Figure 2, data traffic is copied and fed at the source to both the working and the protection entities. The traffic on the working and the protection entities is transmitted simultaneously to the sink of the protection domain, where selection between the working and protection entities is performed (based on some predetermined criteria). |---------------Protection Domain---------------| ============================== /**********Working path************\ +--------+ ============================== +--------+ | Node /| |\ Node | | A {< | | >} Z | | \| |/ | +--------+ ============================== +--------+ \**********Protection path*********/ ============================== Figure 2: 1+1 Protection Architecture Note that control traffic between the edges of the protection domain (such as OAM or a control protocol to coordinate the protection state, etc.) may be transmitted on an entity that differs from the one used for the protected traffic. These packets should not be discarded by the sink.
In 1+1 unidirectional protection switching, there is no need to coordinate the protection state between the protection controllers at both ends of the protection domain. In 1+1 bidirectional protection switching, a protocol is required to coordinate the protection state between the edges of the protection domain. In both protection schemes, traffic flows end-to-end on the working entity after the conditions causing the switchover have been cleared. Data selection may return to selecting traffic from the working entity if reversion is enabled, and it will require coordination of the protection state between the edges of the protection domain. To avoid frequent switching caused by intermittent defects or failures when the network is not stable, traffic is not selected from the working entity before the Wait-To-Restore (WTR) timer has expired.4.7.3. P2MP Linear Protection
Linear protection may be applied to protect unidirectional P2MP entities using 1+1 protection architecture. The source/root MPLS-TP node bridges the user traffic to both the working and protection entities. Each sink/leaf MPLS-TP node selects the traffic from one entity according to some predetermined criteria. Note that when there is a fault condition on one of the branches of the P2MP path, some leaf MPLS-TP nodes may select the working entity, while other leaf MPLS-TP nodes may select traffic from the protection entity. In a 1:1 P2MP protection scheme, the source/root MPLS-TP node needs to identify the existence of a fault condition on any of the branches of the network. This means that the sink/leaf MPLS-TP nodes need to notify the source/root MPLS-TP node of any fault condition. This also necessitates a return path from the sinks/leaves to the source/root MPLS-TP node. When protection switching is triggered, the source/root MPLS-TP node selects the protection transport path for traffic transfer. A form of "segment recovery for P2MP LSPs" could be constructed. Given a P2MP LSP, one can protect any possible point of failure (link or node) using N backup P2MP LSPs. Each backup P2MP LSP originates from the upstream node with respect to a different possible failure point and terminates at all of the destinations downstream of the potential failure point. In case of a failure, traffic is redirected to the backup P2MP path. Note that such mechanisms do not yet exist, and their exact behavior is for further study. A 1:n protection scheme for P2MP transport paths is also required by [RFC5654]. Such a mechanism is for future study.
4.7.4. Triggers for the Linear Protection Switching Action
Protection switching may be performed when: o A defect condition is detected on the working entity, and the protection entity has "no" or an inferior condition. Proactive in-band OAM Continuity Check and Connectivity Verification (CC-V) monitoring of both the working and the protection entities may be used to enable the rapid detection of a fault condition. For protection switching, it is common to run a CC-V every 3.33 ms. In the absence of three consecutive CC-V messages, a fault condition is declared. In order to monitor the working and the protection entities, an OAM Maintenance Entity Group should be defined for each entity. OAM indications associated with fault conditions should be provided at the edges of the protection domain that is responsible for the protection-switching operation. Input from OAM performance monitoring that indicates degradation in the working entity may also be used as a trigger for protection switching. In the case of degradation, switching to the protection entity is needed only if the protection entity can exhibit better operating conditions. o An indication is received from a lower-layer server that there is a defect in the lower layer. o An external operator command is received (e.g., 'Forced Switch', 'Manual Switch'). For details, see Section 6.1.2. o A request to switch over is received from the far end. The far end may initiate this request, for example, on receipt of an administrative request to switch over, or when bidirectional 1:1 protection switching is supported and a defect occurred that could only be detected by the far end, etc. As described above, the protection state should be coordinated between the end points of the protection domain. Control messages should be exchanged between the edges of the protection domain to coordinate the protection state of the edge nodes. Control messages can be delivered using an in-band, data-plane-driven control protocol or a control-plane-based protocol. For 50-ms protection switching, it is recommended that an in-band, data-plane-driven signaling protocol be used in order to coordinate the protection states. An in-band, data-plane protocol for use in MPLS-TP networks is documented in [MPLS-TP-LP] for linear protection (ring protection is discussed in Section 4.8 of this document). This protocol is also used to detect mismatches between the configurations provisioned at the ends of the protection domain.
As described in Section 6.5, the GMPLS control plane already includes procedures and message elements to coordinate the protection states between the edges of the protection domain. These procedures and protocol messages are specified in [RFC4426], [RFC4872], and [RFC4873]. However, these messages lack the capability to coordinate the revertive/non-revertive behavior and the consistency of configured timers at the edges of the protection domain (timers such as WTR, hold-off timer, etc.).4.7.5. Applicability of Linear Protection for LSP Segments
In order to implement data-plane-based linear protection on LSP segments, use is made of the Sub-Path Maintenance Element (SPME), an MPLS-TP architectural element defined in [RFC5921]. Maintenance operations (e.g., monitoring, protection, or management) engage with message transmission (e.g., OAM, Protection Path Coordination, etc.) in the maintained domain. Further discussion of the architecture for OAM and SPME is found in [RFC5921] and [RFC6371]. An SPME is an LSP that is basically defined and used for the purposes of OAM monitoring, protection, or management of LSP segments. The SPME uses the MPLS construct of a hierarchical, nested LSP, as defined in [RFC3031]. For linear protection, SPMEs should be defined over the working and protection entities between the edges of a protection domain. OAM messages and messages used to coordinate protection state can be initiated at the edge of the SPME and sent to the peer edge of the SPME. Note that these messages are sent over the Generic Associated Channel (G-ACh) within the SPME, and that they use a two-label stack, the SPME label, and, at the bottom of the stack, the G-ACh label (GAL) [RFC5586]. The end-to-end traffic of the LSP, which includes data traffic and control traffic (messages for OAM, management, signaling, and to coordinate protection state), is tunneled within the SPMEs by means of label stacking, as defined in [RFC3031]. Mapping between an LSP and an SPME can be 1:1; this is similar to the ITU-T Tandem Connection element that defines a sub-layer corresponding to a segment of a path. Mapping can also be 1:n to allow the scalable protection of a set of LSP segments traversing the part of the network in which a protection domain is defined. Note that each of these LSPs can be initiated or terminated at different end points in the network, but that they all traverse the protection domain and share similar constraints (such as requirements for quality of service (QoS), terms of protection, etc.).
Note also that in the context of segment protection, the SPMEs serve as the working and protection entities.4.7.6. Shared Mesh Protection
For shared mesh protection, the protection resources are used to protect multiple LSPs that do not all share the same end points; for example, in Figure 3 there are two paths, ABCDE and VWXYZ. These paths do not share end points and cannot, therefore, make use of 1:n linear protection, even though they do not have any common points of failure. ABCDE may be protected by the path APQRE, while VWXYZ can be protected by the path VPQRZ. In both cases, 1:1 or 1+1 protection may be used. However, it can be seen that if 1:1 protection is used for both paths, the PQR network segment does not carry traffic when no failures affect either of the two working paths. Furthermore, in the event of only one failure, the PQR segment carries traffic from only one of the working paths. Thus, it is possible for the network resources on the PQR segment to be shared by the two recovery paths. In this way, mesh protection can substantially reduce the number of network resources that have to be reserved in order to provide 1:n protection. A----B----C----D----E \ / \ / \ / P-----Q-----R / \ / \ / \ V----W----X----Y----Z Figure 3: A Shared Mesh Protection Topology As the network becomes more complex and the number of LSPs increases, the potential for shared mesh protection also increases. However, this can quickly become unmanageable owing to the increased complexity. Therefore, shared mesh protection is normally pre- planned and configured by the operator, although an automated system cannot be ruled out. Note that shared mesh protection operates as 1:n linear protection (see Section 4.7.1). However, the protection state needs to be coordinated between a larger number of nodes: the end points of the shared concatenated protection segment (nodes P and R in the example)
as well as the end points of the protected LSPs (nodes A, E, V, and Z in the example). Additionally, note that the shared-protection resources could be used to carry extra traffic. For example, in Figure 4, an LSP JPQRK could be a preemptable LSP that constitutes extra traffic over the PQR hops; it would be displaced in the event of a protection event. In this case, it should be noted that the protection state must also be coordinated with the ends of the extra-traffic LSPs. A----B----C----D----E \ / \ / \ / J-----P-----Q-----R-----K / \ / \ / \ V----W----X----Y----Z Figure 4: Shared Mesh Protection with Extra Traffic4.8. Ring Networks
Several service providers have expressed great interest in the operation of MPLS-TP in ring topologies; they demand a high degree of survivability functionality in these topologies. Various criteria for optimization are considered in ring topologies, such as: 1. Simplification in ring operation in terms of the number of OAM Maintenance Entities that are needed to trigger the recovery actions, the number of recovery elements, the number of management-plane transactions during maintenance operations, etc. 2. Optimization of resource consumption around the ring, such as the number of labels needed for the protection paths that traverse the network, the total bandwidth required in the ring to ensure path protection, etc. (see R91 of [RFC5654]). [RFC5654] introduces a list of requirements for ring protection covering the recovery mechanisms needed to protect traffic in a single ring as well as traffic that traverses more than one ring. Note that configuration and the operation of the recovery mechanisms in a ring must scale well with the number of transport paths, the number of nodes, and the number of ring interconnects.
The requirements for ring protection are fully compatible with the generic requirements for recovery. The architecture and the mechanisms for ring protection are specified in separate documents. These mechanisms need to be evaluated against the requirements specified in [RFC5654], which includes guidance on the principles for the development of new mechanisms.4.9. Recovery in Layered Networks
In multi-layer or multi-regional networking [RFC5212], recovery may be performed at multiple layers or across nested recovery domains. The MPLS-TP recovery mechanism must ensure that the timing of recovery is coordinated in order to avoid race scenarios. This also allows the recovery mechanism of the server layer to fix the problem before recovery takes place in the MPLS-TP layer, or the MPLS-TP layer to perform recovery before a client network. A hold-off timer is required to coordinate recovery timing in multiple layers or across nested recovery domains. Setting this configurable timer involves a trade-off between rapid recovery and the creation of a race condition where multiple layers respond to the same fault, potentially allocating resources in an inefficient manner. Thus, the detection of a defect condition in the MPLS-TP layer should not immediately trigger the recovery process if the hold-off timer is configured as a value other than zero. Instead, the hold-off timer should be started when the defect is detected and, on expiry, the recovery element should be checked to determine whether the defect condition still exists. If it does exist, the defect triggers the recovery operation. The hold-off timer should be configurable. In other configurations, where the lower layer does not have a restoration capability, or where it is not expected to provide protection, the lower layer needs to trigger the higher layer to immediately perform recovery. Although this can be forced by configuring the hold-off timer as zero, it may be that because of layer independence, the higher layer does not know whether the lower layer will perform restoration. In this case, the higher layer will configure a non-zero hold-off timer and rely on the receipt of a specific notification from the lower layer if the lower layer cannot perform restoration. Since layer boundaries are always within nodes, such coordination is implementation-specific and does not need to be covered here.
Reference should be made to [RFC3386], which discusses the interaction between layers in survivable networks.4.9.1. Inherited Link-Level Protection
Where a link in the MPLS-TP network is formed through connectivity (i.e., a packet or non-packet LSP) in a lower-layer network, that connectivity may itself be protected; for example, the LSP in the lower-layer network may be provisioned with 1+1 protection. In this case, the link in the MPLS-TP network has an inherited grade of protection. An LSP in the MPLS-TP network may be provisioned with protection in the MPLS-TP network, as already described, or it may be provisioned to utilize only those links that have inherited protection. By classifying the links in the MPLS-TP network according to the grade of protection that they inherited from the server network, it is possible to compute an end-to-end path in the MPLS-TP network that uses only those links with a specific or superior grade of inherited protection. This means that the end-to-end MPLS-TP LSP can be protected at the grade necessary to conform to the SLA without needing to provide any additional protection in the MPLS-TP layer. This reduces complexity, saves network resources, and eliminates protection-switching coordination problems. When the requisite grade of inherited protection is not available on all segments along the path in the MPLS-TP network, segment protection may be used to achieve the desired protection grade. It should be noted, however, that inherited protection only applies to links. Nodes cannot be protected in this way. An operator will need to perform an analysis of the relative likelihood and consequences of node failure if this approach is taken without providing protection in the MPLS-TP LSP or PW layer to handle node failure.4.9.2. Shared Risk Groups
When an MPLS-TP protection scheme is established, it is important that the working and protection paths do not share resources in the network. If this is not achieved, a single defect may affect both the working and the protection paths with the result that traffic cannot be delivered -- since under such a condition the traffic was not protected.
Note that this restriction does not apply to restoration, since this takes place after the fault has occurred, which means that the point of failure can be avoided if an available path exists. When planning a recovery scheme, it is possible to use a topology map of the MPLS-TP layer to select paths that use diverse links and nodes within the MPLS-TP network. However, this does not guarantee that the paths are truly diverse; for example, two separate links in an MPLS-TP network may be provided by two lambdas in the same optical fiber, or by two fibers that cross the same bridge. Moreover, two completely separate MPLS-TP nodes might be situated in the same building with a shared power supply. Thus, in order to achieve proper recovery planning, the MPLS-TP network must have an understanding of the groups of lower-layer resources that share a common risk of failure. From this, MPLS-TP shared risk groups can be constructed that show which MPLS-TP resources share a common risk of failure. Diversity of working and protection paths can be planned, not only with regard to nodes and links but also in order to refrain from using resources from the same shared risk groups.4.9.3. Fault Correlation
In a layered network, a low-layer fault may be detected and reported by multiple layers and may sometimes lead to the generation of multiple fault reports from the same layer. For example, a failure of a data link may be reported by the line cards in an MPLS-TP node, but it could also be detected and reported by the MPLS-TP OAM. Section 4.6 explains how it is important to coordinate the survivability actions configured and operated in a multi-layer network in a way that will avoid over-equipping the survivability resources in the network, while ensuring that recovery actions are performed in only one layer at a time. Fault correlation is about understanding which single event has generated a set of fault reports, so that recovery actions can be coordinated, and so that the fault logging system does not become overloaded. Fault correlation depends on understanding resource use at lower layers, shared risk groups, and a wider view with regard to the way in which the layers are interrelated. Fault correlation is most easily performed at the point of fault detection; for example, an MPLS-TP node that receives a fault notification from the lower layer, and detects a fault on an LSP in the MPLS-TP layer, can easily correlate these two events. Furthermore, if the same node detects multiple faults on LSPs that
share the same faulty data link, it can easily correlate them. Such a node may use correlation to perform group-based recovery actions and can reduce the number of alarm events that it generates to its management station. Fault correlation may also be performed at a management station that receives fault reports from different layers and different nodes in the network. This enables the management station to coordinate management-originated recovery actions and to present consolidated fault information to the user and automated management systems. It is also necessary to correlate fault information detected and reported through OAM. This function would enable a fault detected at a lower layer, and reported at a transit node of an MPLS-TP LSP, to be correlated with an MPLS-TP-layer fault detected at a Maintenance End Point (MEP) -- for example, the egress of the MPLS-TP LSP. Such correlation allows the coordination of recovery actions performed at the MEP, but it also requires that the lower-layer fault information is propagated to the MEP, which is most easily achieved using a control plane, management plane, or OAM message.
(page 37 continued on part 3)