Applicability of a Stateful Path Computation Element (PCE)

In the following sections, several use cases are described, showcasing scenarios that benefit from the deployment of a stateful PCE.

The following use cases demonstrate a need for visibility into global LSP states in PCE path computations, and for a PCE control of sequence and timing in altering LSP path characteristics within and across PCEP sessions. Reference topologies for the use cases described later in this section are shown in Figures and . Some of the use cases below are focused on MPLS-TE deployments, but may also apply to GMPLS. Unless otherwise cited, use cases assume that all LSPs listed exist at the same LSP priority. The main benefit in the cases below comes from moving away from an asynchronous PCC-driven mode of operation to a model that allows for central control over LSP computations and maintenance, and focuses specifically on the active stateful PCE model of operation.

Because LSP attribute changes in are driven by Path Computation Request (PCReq) messages under control of a PCC's local timers, the sequence of resource reservation arrivals occurring in the network will be randomized. This, coupled with a lack of global LSP state visibility on the part of a stateless PCE may result in suboptimal throughput in a given network topology, as will be shown in the example below. Reference topology 2 in and Tables and show an example in which throughput is at 50% of optimal as a result of lack of visibility and synchronized control across PCC's. In this scenario, the decision must be made as to whether to route any portion of the E-G demand, as any demand routed for this source and destination will decrease system throughput. Link Metric Capacity A-E110 B-F110 C-G110 E-F110 F-G110 Time LSP Src Dst Demand Routable Path 11EG10YesE-F-G 22AB10No--- 31FC10No--- In many cases throughput maximization becomes a bin packing problem. While bin packing itself is an NP-hard problem, a number of common heuristics which run in polynomial time can provide significant improvements in throughput over random reservation event distribution, especially when traversing links which are members of the minimum cut set for a large subset of source destination pairs. Tables and show a simple use case using Reference Topology 1 in , where LSP state visibility and control of reservation order across PCCs would result in significant improvement in total throughput. Link Metric Capacity A-C110 B-C110 C-E105 C-D110 D-E110 Time LSP Src Dst Demand Routable Path 11AE5YesA-C-D-E 22BE10No---

This section discusses a use case of cross-LSP impact under degraded operation. Most existing RSVP-TE implementations will not tear down established LSPs in the event of the failure of the bandwidth increase procedure detailed in . This behavior is directly implied to be correct in and is often desirable from an operator's perspective, because either a) the destination prefixes are not reachable via any means other than MPLS or b) this would result in significant packet loss as demand is shifted to other LSPs in the overlay mesh. In addition, there are currently few implementations offering dynamic ingress admission control (policing of the traffic volume mapped onto an LSP) at the label edge router (LER). Having ingress admission control on a per LSP basis is not necessarily desirable from an operational perspective, as a) one must over-provision tunnels significantly in order to avoid deleterious effects resulting from stacked transport and flow control systems (for example for tunnels that are dynamically resized based on current traffic) and b) there is currently no efficient commonly available northbound interface for dynamic configuration of per LSP ingress admission control. Lack of ingress admission control coupled with the behavior in may result in LSPs operating out of profile for significant periods of time. It is reasonable to expect that these out-of-profile LSPs will be operating in a degraded state and experience traffic loss, but because they end up sharing common network interfaces with other LSPs operating within their bandwidth reservations, thus impacting the operation of the in-profile LSPs, even when there is unused network capacity elsewhere in the network. Furthermore, this behavior will cause information loss in the TED with regards to the actual available bandwidth on the links used by the out-of-profile LSPs, as the reservations on the links no longer reflect the capacity used. Reference Topology 1 in and Tables and show a use case that demonstrates this behavior. Two LSPs, LSP 1 and LSP 2 are signaled with demand 2 and routed along paths A-C-D-E and B-C-D-E respectively. At a later time, the demand of LSP 1 increases to 20. Under such a demand, the LSP cannot be resignaled. However, the existing LSP will not be torn down. In the absence of ingress policing, traffic on LSP 1 will cause degradation for traffic of LSP 2 (due to oversubscription on the links C-D and D-E), as well as information loss in the TED with regard to the actual network state. The problem could be easily ameliorated by global visibility of LSP state coupled with PCC-external demand measurements and placement of two LSPs on disjoint links. Note that while the demand of 20 for LSP 1 could never be satisfied in the given topology, what could be achieved would be isolation from the ill-effects of the (unsatisfiable) increased demand. Link Metric Capacity A-C110 B-C110 C-E105 C-D110 D-E110 Time LSP Src Dst Demand Routable Path 11AE2YesA-C-D-E 22BE2YesB-C-D-E 31AE20No---

As a result of both the lack of visibility into global LSP state and the lack of control over event ordering across PCE sessions, unnecessary perturbations may be introduced into the network by a stateless PCE. Tables and show an example of an unnecessary network perturbation using Reference Topology 1 in . In this case an unimportant (high LSP priority value) LSP (LSP1) is first set up along the shortest path. At time 2, which is assumed to be relatively close to time 1, a second more important (lower LSP-priority value) LSP (LSP2) is established, preempting LSP1, potentially causing traffic loss. LSP1 is then reestablished on the longer A-C-E path. Link Metric Capacity A-C110 B-C110 C-E1010 C-D110 D-E110 Time LSP Src Dst Demand LSP Prio Routable Path 11AE77YesA-C-D-E 22BE70YesB-C-D-E 31AE77YesA-C-E A stateful PCE can help in this scenario by computing both routes at the same time. The advantages of using a stateful PCE over exploiting a stateless PCE via Global Concurrent Optimization(GCO) are three folds. First is the ability to accommodate concurrent path computation from different PCCs. Second is the reduction of control plane overhead since the stateful PCE has the route information of the affected LSPs. Thirdly, the stateful PCE can use the LSP-DB to further optimize the placement of LSPs. This will ensure placement of the more important LSP along the shortest path, avoiding the setup and subsequent preemption of the lower priority LSP. Similarly, when a new higher priority LSP which requires preemption of existing lower priority LSP(s), a stateful PCE can determine the minimum number of lower priority LSP(s) to reroute using the make-before-break (MBB) mechanism without disrupting any service and then set up the higher priority LSP.

Randomization of reservation events caused by lack of control over event ordering across PCE sessions results in poor predictability in LSP routing. An offline system applying a consistent optimization method will produce predictable results to within either the boundary of forecast error (when reservations are over-provisioned by reasonable margins) or to the variability of the signal and the forecast error (when applying some hysteresis in order to minimize churn). Predictable results are valuable for being able to simulate the network and reliably test it under various scenarios, especially under various failure modes and planned maintenances when predictable path characteristics are desired under contention for network resources. Reference Topology 1 and Tables , and show the impact of event ordering and predictability of LSP routing. Link Metric Capacity A-C110 B-C110 C-E110 C-D110 D-E110 Time LSP Src Dst Demand Routable Path 11AE7YesA-C-E 22BE7YesB-C-D-E Time LSP Src Dst Demand Routable Path 12BE7YesB-C-E 21AE7YesA-C-D-E As can be shown in the example, both LSPs are routed in both cases, but along very different paths. This would be a challenge if reliable simulation of the network is attempted. A stateful PCE can solve this through control over LSP ordering.

The bandwidth requirement of LSPs often change over time, requiring resizing the LSP. In most implementations available today, the head-end node performs this function by monitoring the actual bandwidth usage, triggering a recomputation and resignaling when a threshold is reached. This operation is referred as auto-bandwidth adjustment. The head-end node either recomputes the path locally, or it requests a recomputation from a PCE by sending a PCReq message. In the latter case, the PCE computes a new path and provides the new route suggestion. Upon receiving the reply from the PCE, the PCC re-signals the LSP in Shared-Explicit (SE) mode along the newly computed path. With a stateless PCE, the head-end node needs to provide the current used bandwidth and the route information via path computation request messages. Note that in this scenario, the head-end node is the one that drives the LSP resizing based on local information, and that the difference between using a stateless and a passive stateful PCE is in the level of optimization of the LSP placement as discussed in the previous section. A more interesting smart bandwidth adjustment case is one where the LSP resizing decision is done by an external entity, with access to additional information such as historical trending data, application-specific information about expected demands or policy information, as well as knowledge of the actual desired flow volumes. In this case an active stateful PCE provides an advantage in both the computation with knowledge of all LSPs in the domain and in the ability to trigger bandwidth modification of the LSP.

Bandwidth scheduling allows network operators to reserve resources in advance according to the agreements with their customers, and allow them to transmit data with specified starting time and duration, for example for a scheduled bulk data replication between data centers. Traditionally, this can be supported by network management system (NMS) operation through path pre-establishment and activation on the agreed starting time. However, this does not provide efficient network usage since the established paths exclude the possibility of being used by other services even when they are not used for undertaking any service. It can also be accomplished through GMPLS protocol extensions by carrying the related request information (e.g., starting time and duration) across the network. Nevertheless, this method inevitably increases the complexity of signaling and routing process. A passive stateful PCE can support this application with better efficiency since it can alleviate the burden of processing on network elements. This requires the PCE to maintain the scheduled LSPs and their associated resource usage, as well as the ability of head-ends to trigger signaling for LSP setup/deletion at the correct time. This approach requires coarse time synchronization between PCEs and PCCs. If an active stateful PCE is available, the PCE can trigger the setup/deletion of scheduled requests in a centralized manner, as specified , without modification of existing head-end behaviors, by notifying the PCCs to set up or tear down the paths.

The recovery use cases discussed in the following sections show how leveraging a stateful PCE can simplify the computation of recovery path(s). In particular, two characteristics of a stateful PCE are used: 1) using information stored in the LSP-DB for determining shared protection resources and 2) performing computations with knowledge of all LSPs in a domain.

If a PCC can specify in a request whether the computation is for a working path or for protection, and a PCC can report the resource as a working or protection path, then the following text applies. A PCC can send multiple requests to the PCE, asking for two LSPs and use them as working and backup paths separately. Either way, the resources bound to backup paths can be shared by different LSPs to improve the overall network efficiency, such as m:n protection or pre-configured shared mesh recovery techniques as specified in . If resource sharing is supported for LSP protection, the information relating to existing LSPs is required to avoid allocation of shared protection resources to two LSPs that might fail together and cause protection contention issues. A stateless PCE can accommodate this use case by having the PCC pass this information as a constraint in the path computation request. A passive stateful PCE can more easily accommodate this need using the information stored in its LSP-DB. Furthermore, an active stateful PCE can help with (re)-optimizization of protection resource sharing as well as LSP maintenance operation with fewer impact on protection resources.

For example, in the network depicted in Figure , suppose there exists LSP1 with working path LSP1_working following A->E and with backup path LSP1_backup following A->B->E. A request arrives asking for a working and backup path pair to be computed for LSP2 from B to E. If the PCE decides LSP2_working follows B->A->E, then the backup path LSP2_backup should not share the same protection resource with LSP1 since LSP2 shares part of its resource (specifically A->E) with LSP1 (i.e., these two LSPs are in the same shared risk group). There is no such constraint if B->C->D->E is chosen for LSP2_working. If a stateless PCE is used, the head node B needs to be aware of the existence of LSPs which share the route of LSP2_working and of the details of their protection resources. B must pass this information to the PCE as a constraint so as to request a path with diversity. Alternatively, a stateless PCE may able to compute Shared Risk Link Group (SRLG)-diversified paths if TED is extended so that it includes the SRLG information that are protected by a given backup resource, but at the expense of a high complexity in routing. On the other hand, a stateful PCE can get the LSPs information by itself given that the LSP identifier(s) and can achieve the goal of finding SRLG-diversified protection paths for both LSPs. This is made possible by comparing the LSP resource usage exploiting the LSP-DB accessible by the stateful PCE.

In case of a link failure, such as a fiber cut, multiple LSPs may fail at the same time. Thus, the source nodes of the affected LSPs will be informed of the failure by the nodes detecting the failure. These source nodes will send requests to a PCE for rerouting. In order to reuse the resource taken by an existing LSP, the source node can send a PCReq message including the Exclude Route Object (XRO) with Fail (F) bit set, together with the record route object (RRO) containing the current route information, as specified in . If a stateless PCE is used, it might respond to the rerouting requests separately if they arrive at different times. Thus, it might result in sub-optimal resource usage. Even worse, it might unnecessarily block some of the rerouting requests due to insufficient resources for later-arrived rerouting messages. If a passive stateful PCE is used to fulfill this task, the procedure can be simplified. The PCCs reporting the failures can include LSP identifiers instead of detailed information and the PCE can find relevant LSP information by inspecting the LSP-DB. Moreover, the PCE can re-compute the affected LSPs concurrently while reusing part of the existing LSPs resources when it is informed of the failed link identifier provided by the first request. This is made possible since the passive stateful PCE can check what other LSPs are affected by the failed link and their route information by inspecting its LSP-DB. As a result, a better performance can be achieved, such as better resource usage or minimal probability of blocking upcoming new rerouting requests sent as a result of the link failure. If the target is to avoid resource contention within the time-window of high number of LSP rerouting requests, a stateful PCE can retain the under-construction LSP resource usage information for a given time and exclude it from being used for forthcoming LSPs request. In this way, it can ensure that the resource will not be double-booked and thus the issue of resource contention and computation crank-backs can be alleviated.

An alternative way to achieve efficient resilience is to maintain SRLG disjointness between LSPs, irrespective of whether these LSPs share the source and destination nodes or not. This can be achieved at provisioning time, if the routes of all the LSPs are requested together, using a synchronized computation of the different LSPs with SRLG disjointness constraint. If the LSPs need to be provisioned at different times, the PCC can specify, as constraints to the path computation a set of SRLGs using the Exclude Route Object . However, for the latter to be effective, it is needed that the entity that requests the route to the PCE maintains updated SRLG information of all the LSPs to which it must maintain the disjointness. A stateless PCE can compute an SRLG-disjoint path by inspecting the TED and precluding the links with the same SRLG values specified in the PCReq message sent by a PCC. A passive stateful PCE maintains the updated SRLG information of the established LSPs in a centralized manner. Therefore, the PCC can specify as constraints to the path computation the SRLG disjointness of a set of already established LSPs by only providing the LSP identifiers. Similarly, a passive stateful PCE can also accommodate disjointness using other constraints, such as link, node or path segment etc.

In Multi-Layer Networks (MLN), a Virtual Network Topology (VNT) consists of a set of one or more TE LSPs in the lower layer which provides TE links to the upper layer. In , the PCE-based architecture is proposed to support path computation in MLN networks in order to achieve inter-layer TE. The establishment/teardown of a TE link in VNT needs to take into consideration the state of existing LSPs and/or new LSP request(s) in the higher layer. Hence, when a stateless PCE cannot find the route for a request based on the upper layer topology information, it does not have enough information to decide whether to set up or remove a TE link or not, which then can result in non-optimal usage of resource. On the other hand, a passive stateful PCE can make a better decision of when and how to modify the VNT either to accommodate new LSP requests or to re-optimize resource usage across layers irrespective of the PCE models as described in . Furthermore, given the active capability, the stateful PCE can issue VNT modification suggestions in order to accommodate path setup requests or re-optimize resource usage across layers.

In order to make efficient usage of network resources, it is sometimes desirable to re-optimize one or more LSPs dynamically. In the case of a stateless PCE, in order to optimize network resource usage dynamically through online planning, a PCC must send a request to the PCE together with detailed path/bandwidth information of the LSPs that need to be concurrently optimized. This means the PCC must be able to determine when and which LSPs should be optimized. In the case of a passive stateful PCE, given the LSP state information in the LSP database, the process of dynamic optimization of network resources can be simplified without requiring the PCC to supply detailed LSP state information. Moreover, an active stateful PCE can even make the process automated by triggering the request since a stateful PCE can maintain information for all LSPs that are in the process of being set up and it may have the ability to control timing and sequence of LSP setup/deletion, the optimization procedures can be performed more intelligently and effectively. A stateful PCE can also determine which LSP should be re-optimized based on network events. For example, when a LSP is torn down, its resources are freed. This can trigger the stateful PCE to automatically determine which LSP should be reoptimized so that the recently freed resources may be allocated to it. A special case of LSP re-optimization is GCO . Global control of LSP operation sequence in is predicated on the use of what is effectively a stateful (or semi-stateful) NMS. The NMS can be either not local to the network nodes, in which case another northbound interface is required for LSP attribute changes, or local/collocated, in which case there are significant issues with efficiency in resource usage. A stateful PCE adds a few features that: Roll the NMS visibility into the PCE and remove the requirement for an additional northbound interface Allow the PCE to determine when re-optimization is needed, with which level (GCO or a more incremental optimization) Allow the PCE to determine which LSPs should be re-optimized Allow a PCE to control the sequence of events across multiple PCCs, allowing for bulk (and truly global) optimization, LSP shuffling etc.

If LSPs are dynamically allocated and released over time, the resource becomes fragmented. In networks with link bundle, the overall available resource on a (bundle) link might be sufficient for a new LSP request, but if the available resource is not continuous, the request is rejected. In order to perform the defragmentation procedure, stateful PCEs can be used, since global visibility of LSPs in the network is required to accurately assess resources on the LSPs, and perform de-fragmentation while ensuring a minimal disruption of the network. This use case cannot be accommodated by a stateless PCE since it does not possess the detailed information of existing LSPs in the network. Another case of particular interest is the optical spectrum defragmentation in flexible grid networks. In Flexible grid networks , LSPs with different optical spectrum sizes (such as 12.5GHz, 25GHz etc.) can co-exist so as to accommodate the services with different bandwidth requests. Therefore, even if the overall spectrum size can meet the service request, it may not be usable if the available spectrum resource is not contiguous, but rather fragmented into smaller pieces. Thus, with the help of existing LSP state information, a stateful PCE can make the resource grouped together to be usable. Moreover, a stateful PCE can proactively choose routes for upcoming path requests to reduce the chance of spectrum fragmentation.

PCE has been identified as an appropriate technology for the determination of the paths of point-to-multipoint (P2MP) TE LSPs . The application scenarios and use-cases described in , and are also applicable to P2MP TE LSPs. In addition to these, the stateful nature of a PCE simplifies the information conveyed in PCEP messages since it is possible to refer to the LSPs via an identifier. For P2MP, this is an added advantage, where the size of the PCEP message is much larger. In case of stateless PCEs, modification of a P2MP tree requires encoding of all leaves along with the paths in PCReq message. But using a stateful PCE with P2MP capability, the PCEP message can be used to convey only the modifications (the other information can be retrieved from the identifier via the LSP-DB).

In Wavelength Switched Optical Networks (WSONs) , a wavelength-switched LSP traverses one or more fiber links. The bit rates of the client signals carried by the wavelength LSPs may be the same or different. Hence, a fiber link may transmit a number of wavelength LSPs with equal or mixed bit rate signals. For example, a fiber link may multiplex the wavelengths with only 10Gb/s signals, mixed 10Gb/s and 40Gb/s signals, or mixed 40Gb/s and 100Gb/s signals. IA-RWA in WSONs refers to the process (i.e., lightpath computation) that takes into account the optical layer/transmission imperfections by considering as additional (i.e., physical layer) constraints. To be more specific, linear and non-linear effects associated with the optical network elements should be incorporated into the route and wavelength assignment procedure. For example, the physical imperfection can result in the interference of two adjacent lightpaths. Thus, a guard band should be reserved between them to alleviate these effects. The width of the guard band between two adjacent wavelengths depends on their characteristics, such as modulation formats and bit rates. Two adjacent wavelengths with different characteristics (e.g., different bit rates) may need a wider guard band and with same characteristics may need a narrower guard band. For example, 50GHz spacing may be acceptable for two adjacent wavelengths with 40G signals. But for two adjacent wavelengths with different bit rates (e.g., 10G and 40G), a larger spacing such as 300GHz spacing may be needed. Hence, the characteristics (states) of the existing wavelength LSPs should be considered for a new RWA request in WSON. In summary, when stateful PCEs are used to perform the IA-RWA procedure, they need to know the characteristics of the existing wavelength LSPs. The impairment information relating to existing and to-be-established LSPs can be obtained by nodes in WSON networks via external configuration or other means such as monitoring or estimation based on a vendor-specific impair model. However, WSON related routing protocols, i.e., and , only advertise limited information (i.e., availability) of the existing wavelengths, without defining the supported client bit rates. It will incur substantial amount of control plane overhead if routing protocols are extended to support dissemination of the new information relevant for the IA-RWA process. In this scenario, stateful PCE(s) would be a more appropriate mechanism to solve this problem. Stateful PCE(s) can exploit impairment information of LSPs stored in LSP-DB to provide accurate RWA calculation.