]> Ciena Corporation

mkoldych@proton.me

Ciena Corporation

ssivabal@ciena.com

Huawei Technologies

pengshuping@huawei.com

Nokia

diego.achaval@nokia.com

Juniper Networks, Inc

hkotni@juniper.net

Nokia

andrew.stone@nokia.com

Path Computation Element This document clarifies certain operational behavior aspects of the Path Computation Element Communication Protocol (PCEP). The content of this document has been compiled based on several interop exercises. This document does not make any updates or revisions to any PCEP specifications, instead it provides additional information to aid in the interpretation of those documents. Discussion Venues Discussion of this document takes place on the Path Computation Element mailing list (pce@ietf.org), which is archived at . Source for this draft and an issue tracker can be found at .

Introduction Due to different interpretations of the PCEP standards , , , , and , it was found that implementations often had to adjust their behavior in order to interoperate. The current document serves to clarify certain aspects of PCEP to make it easier to produce interoperable implementations of PCEP. This document does not make any updates or revisions to any PCEP specifications. It provides additional informational guidance to aid in the consistent interpretation of those documents across implementations.

Terminology The following terminology is used in this document: PCC: Path Computation Client . PCE: Path Computation Element . PCEP: Path Computation Element Communication Protocol . MBB: Make-Before-Break. A procedure during which the head-end of a traffic-engineered path wishes to move traffic to a new path without losing any traffic, by first "making" a new path and then "breaking" the old path (see ). Association parameters: As described in , the combination of the mandatory fields Association type, Association ID and Association Source in the ASSOCIATION object uniquely identify the association group. If the optional TLVs - Global Association Source or Extended Association ID are included, then they are included in combination with mandatory fields to uniquely identify the association group. Association information: As described in , the ASSOCIATION object could also include other optional TLVs based on the association types, that provides 'information' related to the association type. ERO: Explicit Route Object is the path of the LSP encoded into a PCEP object. This document distinguishes three cases for the ERO object, because an absent ERO and an empty ERO are not equivalent:

• An absent ERO object, i.e., one not included in the message at all.
• An empty ERO object, i.e., one that is present but carries no subobjects, is represented with the notation "ERO={}".
• An ERO object containing a given sequence of subobjects is represented as "ERO={A}".

PCRPT-LSP-DB: PCEP Reported Label Switched Path Database. A logical datastore that captures the reported state of Label Switched Paths (LSPs), organized as Tunnels and their LSPs. Each PCEP speaker maintains its own local copy, populated from PCRpt messages. On a PCE, the PCRPT-LSP-DB holds the state reported by all PCCs that report to that PCE. On a PCC, it holds the state for the LSPs for which that PCC is the head-end. This term is not defined in the PCE architecture; however, it is used in this document to describe how a PCEP speaker maintains LSP-related state reported via PCRpt messages. EXTENDED-LSP-DB: Extended Label Switched Path Database. A single, optional, implementation-specific logical datastore that holds entries for multiple LSPs, keyed using the same identifiers as the PCRPT-LSP-DB. It is not required by a conforming PCEP implementation. It is typically maintained on the PCE to hold additional attributes such as desired (intended) state, telemetry data, operator configuration, and other information not defined within IETF PCE working group documents; a PCC may maintain an equivalent datastore where useful. This term is not defined in the PCE architecture and is used in this document only to refer to this conceptual datastore. PCE-ASSO-DB: PCEP Association Database. A logical datastore that captures PCEP Association membership, populated by PCRpt messages and/or by configuration on the PCE. This term is not defined in the PCE architecture; it is used in this document to describe how a PCE maintains Association state. See the PCEP Association Database section. PLSP-ID (Path LSP Identifier): Introduced in . A unique identifier used in PCEP to distinguish a specific LSP between a PCC and a PCE which is constant for the lifetime of a PCEP session.

PCEP LSP Database This document uses the concept of the PCRPT-LSP-DB, a database of actual LSP state in the network, to illustrate the internal state of PCEP speakers in response to PCRpt messages. Note that the term "LSP", which stands for "Label Switched Path", if taken too literally, would restrict the discussion to the MPLS dataplane only. In this document, the term "LSP" is applied to non-MPLS paths as well, to avoid renaming the term.

Structure The distinction between Tunnels and LSPs in the PCRPT-LSP-DB is essential for accurately modeling state information in PCEP, including procedures such as make-before-break, and for maintaining consistent state management across PCEP speakers. The PCRPT-LSP-DB contains two types of information: LSPs and Tunnels. An LSP is identified by the LSP-IDENTIFIERS TLV, while a Tunnel is identified by the PLSP-ID in the LSP object and/or the SYMBOLIC-NAME (see ). A Tunnel in PCEP does not necessarily map one-to-one to a single tunnel construct on the router. A tunnel into which a client inserts traffic may be supported by one or more underlying transport paths. For example, the working and protect paths of a single tunnel interface on the router are represented in PCEP as two different Tunnels with different PLSP-IDs. An LSP can be viewed as an instance of a Tunnel. In steady state, a Tunnel typically has a single LSP; however, during make-before-break procedures (see ), multiple LSPs may exist simultaneously to represent both the new and old instances during the transition.

Synchronization Since the PCRPT-LSP-DB contains PCEP-specific information such as PLSP-IDs, which remain constant for the duration of a PCEP session, both the PCE and PCC maintain their own local copies of the PCRPT-LSP-DB. The PCE PCRPT-LSP-DB is only modified by PCRpt messages, no other PCEP message may modify the PCE PCRPT-LSP-DB. The PCC PCRPT-LSP-DB is built from actual forwarding state that PCC has installed. PCC uses PCRpt messages to synchronize its local PCRPT-LSP-DB to the PCE. The PCE is intended to act based on the most recent state available in the PCRPT-LSP-DB, which represents the reported state of Label Switched Paths (LSPs) in the network. This does not preclude the PCE from leveraging information obtained outside the PCRPT-LSP-DB. For example, implementation-specific mechanisms may be used to collect traffic statistics that can be considered during path computation. Such additional information, including desired (intended) state, telemetry, and operator configuration, is not part of the PCRPT-LSP-DB itself and may instead be held in the separate, optional EXTENDED-LSP-DB described in the Terminology section. The PCRPT-LSP-DB reflects the reported state of LSPs and does not imply alignment with the desired state. It also does not necessarily reflect the live state of the network, because there are two convergence windows during which the databases can lag the network:

• The network state has changed, but the PCC has not yet updated its own PCRPT-LSP-DB.
• The PCC has updated its PCRPT-LSP-DB, but the PCRpt carrying that change has not yet been received and processed by the PCE.

For example, in the case of a PCE-initiated LSP, a change in the LSP configuration made by an operator represents a modification to the desired state. However, the actual state does not change until the PCE sends a PCInit or PCUpd message to the PCC. Upon receipt of such a message, the PCC may act on the request, update its local PCRPT-LSP-DB, and generate a PCRpt message to inform the PCE of the change. The PCE then updates its own PCRPT-LSP-DB accordingly. Once this exchange is complete, the PCRPT-LSP-DBs on both the PCC and the PCE are synchronized.

Make before Break (MBB) Due to variations in how different implementations interpret or handle MBB procedures, sometimes resulting in incorrect message processing or misinterpretation, the following section provides illustrative examples to clarify expected MBB behavior.

Successful MBB Below is an example of performing MBB to switch a Tunnel from one path to another. The path encoded into the ERO object is represented as ERO={A} and ERO={B}. PCC has an existing LSP in UP state, with LSP-ID=2. PCC sends PCRpt(R-FLAG=0, PLSP-ID=100, LSP-ID=2, ERO={A}, OPER-FLAG=UP). The resulting content of the PCRPT-LSP-DB is shown in Figure 1. PCC initiates the MBB procedure by creating a new LSP with LSP-ID=3. It does not matter what triggered the creation of the new LSP, it could have been due to a new path received via PCUpd (if the given Tunnel is delegated), or it could have been local computation on the PCC (if the Tunnel is locally computed on the PCC), or it could have been a change in configuration on the PCC (if the Tunnel's path is explicitly configured on the PCC). It is important to emphasize that the procedure for updating the PCRPT-LSP-DB is common, regardless of the trigger that caused the change. PCC sends PCRpt(R-FLAG=0, PLSP-ID=100, LSP-ID=3, ERO={B}, OPER- FLAG=UP). Both LSPs now exist simultaneously, as shown in Figure 2. After traffic has successfully switched to the new LSP, the PCC cleans up the old LSP. PCC sends PCRpt(R-FLAG=1, PLSP-ID=100, LSP- ID=2). Only the new LSP remains, as shown in Figure 3.

Aborted MBB The MBB process can abort when the newly created LSP is destroyed before it is installed as traffic carrying. This scenario is described below. PCC has an existing LSP in UP state, with LSP-ID=2. PCC sends PCRpt(R-FLAG=0, OPER-FLAG=UP, PLSP-ID=100, LSP-ID=2). The initial content of the PCRPT-LSP-DB is shown in Figure 4. MBB procedure is initiated, a new LSP is created with LSP-ID=3. LSP is currently being established, so its oper state is DOWN. PCC sends PCRpt(R-FLAG=0, OPER-FLAG=DOWN, PLSP-ID=100, LSP-ID=3). The new LSP is being established while the old LSP remains UP, as shown in Figure 5. MBB procedure is aborted. PCC sends PCRpt(R-FLAG=1, PLSP-ID=100, LSP-ID=3). The new LSP has been removed and only the original LSP remains, as shown in Figure 6.

PCEP Association Database A PCEP Association is the instantiation of a group containing at least one LSP. The PCE-ASSO-DB is populated by PCRpt messages and/or via configuration on the PCE itself. An Association is identified by the Association Parameters. As the Association Parameters contain many fields, all fields are grouped into a single value for convenience. The notation ASSO_PARAM=A and ASSO_PARAM=B is used to refer to different PCEP Associations: A and B, respectively. The message parameter notation used in the examples below (for example, R-FLAG, OPER-FLAG, ASSO_R_FLAG, and ASSO_PARAM) refers to the corresponding flags and fields of the LSP and ASSOCIATION objects defined in and .

LSPs in same Association The following example illustrates how LSPs join the same Association. The PCC creates the first LSP. The PCC sends PCRpt(R-FLAG=0, PLSP-ID=100, LSP-ID=1, ASSO_PARAM=A, ASSO_R_FLAG=0). The resulting content of the PCE-ASSO-DB is shown in Figure 7. The PCC creates the second LSP. The PCC sends PCRpt(R-FLAG=0, PLSP-ID=200, LSP-ID=1, ASSO_PARAM=A, ASSO_R_FLAG=0). Both LSPs are now members of the same Association, as shown in Figure 8. The PCC updates the first LSP. As indicates, subsequent PCRpt should include only the associations that are being modified or removed. Therefore it is optional whether to send the ASSOCIATION object in this PCRpt, since the LSP is already in the Association. The PCC sends PCRpt(R-FLAG=0, PLSP-ID=100, LSP-ID=1). The content of the PCE-ASSO-DB is unchanged. Note that the PCC sends the ASSOCIATION OBJECT in the first PCRpt during SYNC state, even if it has already issued a PCRpt with the association object sometime in the past with this PCE. The synchronization steps outlined in are to be followed. The content of the PCE-ASSO-DB is unchanged, as shown in Figure 9. The PCC decides to delete the second LSP. The PCC sends PCRpt(R-FLAG=1, PLSP-ID=200, LSP-ID=1). Only the first LSP remains in the Association, as shown in Figure 10. The PCC decides to remove the first LSP from the Association, but not delete the LSP itself. The PCC sends PCRpt(R-FLAG=0, PLSP-ID=100, LSP- ID=1, ASSO_PARAM=A, ASSO_R_FLAG=1). The PCE-ASSO-DB is now empty. The Association is now empty, as shown in Figure 11.

Switch Association during MBB The following example illustrates how a Tunnel goes through MBB and switches from Association A to Association B. Each new LSP (identified by the LSP-ID) does not inherit the Association membership of any previous LSPs within the same Tunnel. This is so that a Tunnel can have two LSPs that are in different Associations, this may be done when switching from one Association to another. The PCC creates the first LSP. The PCC sends PCRpt(R-FLAG=0, PLSP-ID=100, LSP-ID=1, ASSO_PARAM=A, ASSO_R_FLAG=0). The resulting content of the PCE-ASSO-DB is shown in Figure 12. The PCC creates the MBB LSP in a different Association. The PCC sends PCRpt(R-FLAG=0, PLSP-ID=100, LSP-ID=2, ASSO_PARAM=B, ASSO_R_FLAG=0). The two LSPs of the Tunnel are now in different Associations, as shown in Figure 13. The PCC deletes the old LSP. The PCC sends PCRpt(R-FLAG=1, PLSP-ID=100, LSP- ID=1). Only the new LSP in Association B remains, as shown in Figure 14.

Computation Constraints As well as reporting any state change in the network on a PCRpt message, a PCC may also change the parameters of a delegated LSP. For example, it may remove or modify the computation constraints that it wishes the PCE to apply as it computes any updated paths in the future. For any PCEP object that specifies a path computation constraint and that does not have a defined explicit removal flag, the absence of that entire object on a repeat or follow-up PCRpt message indicates removal of the constraint previously specified by that object. For example, suppose the first PCRpt contains an LSPA object with some affinity constraints. Then if a subsequent PCRpt does not contain an LSPA object, then this means that the previously specified affinity constraints do not apply anymore. Same applies to all PCEP objects, like METRIC, BANDWIDTH, etc., which do not have an explicit flag for removal. This simply ensures that it is possible to remove a constraint without using an explicit removal flag.

Overloaded PCE defines the concept of an Overloaded PCE and explains how a PCE may signal to a PCC that it is congested, instructing that "no requests should be sent to that PCE until the overload state is cleared." In the case of an overloaded PCE, a PCC implementation could choose to wait for the PCE to no longer be overloaded or instead send a PCReq to a backup, non-overloaded PCE instead. builds upon by introducing the concept of a Stateful PCE, which allows delegation of LSP control to a single PCE. However, it does not address the case of an overloaded PCE. According to , a PCE maintains delegation until it is revoked by the PCC or returned back to PCC by the PCE. The PCC may revoke delegation and re-assign it to another PCE. As a result, a PCE in an overload state still retains LSP delegation. For PCC-initiated LSPs, the PCC may revoke delegation from the overloaded PCE and maintain delegation for itself or delegate it to another PCE. For PCE-initiated LSPs, since the PCC cannot revoke delegation as per , the overloaded PCE may return the delegation to the PCC.

Reports toward an Overloaded PCE The guidance in this section clarifies behavior for reports and does not modify . Any change that would require a PCC to stop sending reports toward an overloaded PCE would be an update to and is out of scope for this document. There is a trade-off in how a PCC handles PCRpt messages while a PCE signals that it is overloaded. Fully reprocessing a flood of PCRpt messages consumes CPU and memory on the PCE, and so continuing to send reports could prolong or worsen the overload condition.However, the longer the PCE is without the latest reported state, the more likely it is to compute paths using stale information. What it means for a PCE to be overloaded is implementation specific. For example, an implementation may use separate resources for ingesting reports and for performing computation, in which case it may be overloaded for computation while still able to process reports. A PCE may also wish to remain synchronized with the network without yet participating in active computation, for example while a new instance is being bootstrapped. Given these trade offs, the following describes the behavior a PCC may adopt toward an overloaded PCE. Sending reports as usual is the baseline interoperable behavior and throttling reports is an optional behavior:

• Sending reports as usual: the PCC continues to send PCRpt messages so that the PCRPT-LSP-DB on the PCE remains synchronized. This is the baseline behavior and requires no change to existing implementations.
• Throttling reports: the PCC may make a local decision about whether to immediately send a PCRpt message to the PCE or to hold that message until the PCE indicates it is no longer overloaded. This decision may depend on how significant the change to the LSP is, how long the message has been held, and how many other messages are held for the same reason. In all cases the PCC should log the situation, applying thresholding to the rate of generation of log events, so that an operator can determine what is happening.

Stopping reports entirely while the PCE is overloaded is not recommended, because it risks the PCRPT-LSP-DB on the PCE going out of sync with the network for the full duration of the overload condition. The decision to throttle cannot depend on local policy alone, because some reports remain significant regardless of policy. At a minimum, the following reports should not be throttled:

• A PCRpt sent in response to a PCUpd, since a PCE may still take actions on a delegated LSP while marked overloaded.
• PCRpt messages sent as part of state synchronization during session establishment.

A PCC may throttle other less significant reports such as frequent or redundant updates, for example local protection availability flags. As an alternative to per-report throttling, a PCE under overload may selectively close one or more PCEP sessions, choosing which sessions to prioritize. When the overload condition clears, the sessions are re-established and standard state synchronization brings the PCE back in sync with the latest state. This avoids the complexity of deciding, on a per-report basis, which reports to hold or drop.

Security Considerations This document does not define any new protocol mechanisms and does not change the PCEP messages or objects defined in , , , or . The security considerations of those documents therefore continue to apply. The clarifications in this document aim to reduce the risk of interoperability problems that arise when implementations interpret PCEP behavior differently. Such divergence can have security relevant consequences such as inconsistent handling of reported state can cause a PCE to compute or install paths based on an inaccurate view of the network, and the absence of an agreed behavior toward an overloaded PCE can be exploited to amplify a denial-of-service condition by sustaining a flood of reports. Conversely, holding or throttling PCRpt messages toward an overloaded PCE, as described in the Overloaded PCE section, can delay the visibility of state changes, including those resulting from misconfiguration or compromise. The logging recommended in that section helps an operator detect such situations. Selective closing of PCEP sessions during overload is an operational control that should be governed by local policy.

Manageability Considerations The logical datastores described in this document (the PCRPT-LSP-DB, the optional EXTENDED-LSP-DB, and the PCE-ASSO-DB) are useful points of operational visibility. Where an implementation exposes them, they map naturally onto state data models, for example YANG. An operator benefits from being able to read these datastores in order to understand the state held by a PCEP speaker. Because the PCRPT-LSP-DB on a PCC and on a PCE can lag one another during the convergence windows described in the Synchronization section, an operator may need to verify that the two are aligned. This can be done, for example, by comparing the set of reported LSPs and their last-reported state, or by observing the state synchronization procedures of . As described in the Overloaded PCE section, a PCC that throttles or holds PCRpt messages toward an overloaded PCE should log that it is doing so, applying thresholding to the rate of log generation, so that an operator can determine what is happening.

IANA Considerations This document has no IANA actions.

Normative References The Path Computation Element Communication Protocol (PCEP) provides mechanisms for Path Computation Elements (PCEs) to perform path computations in response to Path Computation Client (PCC) requests. Although PCEP explicitly makes no assumptions regarding the information available to the PCE, it also makes no provisions for PCE control of timing and sequence of path computations within and across PCEP sessions. This document describes a set of extensions to PCEP to enable stateful control of MPLS-TE and GMPLS Label Switched Paths (LSPs) via PCEP. This document introduces a generic mechanism to create a grouping of Label Switched Paths (LSPs) in the context of a Path Computation Element (PCE). This grouping can then be used to define associations between sets of LSPs or between a set of LSPs and a set of attributes (such as configuration parameters or behaviors), and it is equally applicable to the stateful PCE (active and passive modes) and the stateless PCE. This document describes the use of RSVP (Resource Reservation Protocol), including all the necessary extensions, to establish label-switched paths (LSPs) in MPLS (Multi-Protocol Label Switching). Since the flow along an LSP is completely identified by the label applied at the ingress node of the path, these paths may be treated as tunnels. A key application of LSP tunnels is traffic engineering with MPLS as specified in RFC 2702. [STANDARDS-TRACK] This document specifies the Path Computation Element (PCE) Communication Protocol (PCEP) for communications between a Path Computation Client (PCC) and a PCE, or between two PCEs. Such interactions include path computation requests and path computation replies as well as notifications of specific states related to the use of a PCE in the context of Multiprotocol Label Switching (MPLS) and Generalized MPLS (GMPLS) Traffic Engineering. PCEP is designed to be flexible and extensible so as to easily allow for the addition of further messages and objects, should further requirements be expressed in the future. [STANDARDS-TRACK] The Path Computation Element Communication Protocol (PCEP) provides mechanisms for Path Computation Elements (PCEs) to perform path computations in response to Path Computation Client (PCC) requests. The extensions for stateful PCE provide active control of Multiprotocol Label Switching (MPLS) Traffic Engineering Label Switched Paths (TE LSPs) via PCEP, for a model where the PCC delegates control over one or more locally configured LSPs to the PCE. This document describes the creation and deletion of PCE-initiated LSPs under the stateful PCE model. Constraint-based path computation is a fundamental building block for traffic engineering systems such as Multiprotocol Label Switching (MPLS) and Generalized Multiprotocol Label Switching (GMPLS) networks. Path computation in large, multi-domain, multi-region, or multi-layer networks is complex and may require special computational components and cooperation between the different network domains. This document specifies the architecture for a Path Computation Element (PCE)-based model to address this problem space. This document does not attempt to provide a detailed description of all the architectural components, but rather it describes a set of building blocks for the PCE architecture from which solutions may be constructed. This memo provides information for the Internet community.

Acknowledgments The authors would like to thank Adrian Farrel for a thorough review on this document and Tom Petch and Boris Hassanov for useful review and feedback.

Contributors