PCEP Best Current Practices - Message formats and extensions

The RBNF notation, defined in , is used to specify the message format for the Path Computation Element communication Protocol (PCEP). The core of PCEP has been defined in and later extended, notably, in to support Vendor Extensions; in , adding a CLASSTYPE object to support Diffserv-aware Traffic Engineering (DS-TE); in , for topology confidentiality by means of Path-Keys; in , in support of exclusions; in to convey specific Objective Functions; in , for Global Concurrent Optimization, in , for monitoring and in for point-to-multipoint (P2MP) computation. At the time of writing, several I.-D. are also addressing specific aspects, such as PCEP extensions for GMPLS networks , for hierarchical PCE or for multi-layer, multi-region networks . Stateful PCE capabilites are also being defined in , including the case where a PCE is able to initiate the establishment and release of LSPs in . Most PCEP RFCs describe specific protocol extensions and, as such, they focus on their constructs extending some base RFCs. Although it is not the intention of each individual draft or RFC to provide the latest and most complete/full definition of the protocol messages, in practice combining all the extensions as defined in the respective RFCs is complex, and open to interpretation. Message rules are sometimes provided within the text, resulting in ambiguity. Moreover, the fact that extensions may be defined in parallel may be a problem. The canonical example is the case where RFC X defines construct p ::= A and subsequent RFC Y extends RFC X stating that object C MUST follow object A and RFC Z also extends RFC X stating that object D MUST follow object A. This document describes current practice when implementing existing PCEP RFCs. This involves extending the existing RBNF notations using more verbose constructs where appropiate, while being semantically equivalent, in order to avoid ambiguity and to facilitate message validation.

The use of RBNF states that the ordering of objects and constructs in an assignment is explicit, and protocol specifications MAY opt to state that ordering is only RECOMMENDED (the elements of a list of objects and constructs MAY be received in any order). The core PCEP document states in Section 6 that an implementation MUST form the PCEP messages using the object ordering specified in . equally states that "An implementation MUST form the PCEP messages using the object ordering specified in this document." only states that "the XRO is OPTIONAL and MAY be carried within Path Computation Request (PCReq) and Path Computation Reply (PCRep) messages." and no ordering is provided. For example, it does not mention SVEC objects or rules. specifies that "the OF object MAY be carried within a PCReq message. If an objective function is to be applied to a set of synchronized path computation requests, the OF object MUST be carried just after the corresponding SVEC (Synchronization VECtor) object and MUST NOT be repeated for each elementary request. Similarly, if a metric is to be applied to a set of synchronized requests, the METRIC object MUST follow the SVEC object and MUST NOT be repeated for each elementary request. (...) An OF object specifying an objective function that applies to an individual path computation request (non-synchronized case) MUST follow the RP object for which it applies". It should be understood that this last sentence introduces ambiguity and if interpreted as the OF object MUST strictly follow (right after) the RP object, it contradicts where the RP object is followed by the ENDPOINTS object. RFCs that extend the core PCEP protocol are not consistent with the object ordering. in section 3.2 is not consistent with the ordering of OF and metric-list:

::= [] [] ::= (snip) [] [] ::= [] [] [] [] ]]> In view of the above considerations, this document aims at providing an object ordering for PCEP messages so implementations can interoperate.

PCEP RFCs may use inconsistent or ambiguous naming. For example defines the Open message as having a common header and an OPEN object, and later uses Open to refer to the object that may appear in a PCErr message.

::= ::= ( []) | [] ]]> It is common that a sequence or repetition of an object OBJ is noted as obj-list. It may happen that in extensions to core documents, the naming is kept although it no longer applies to such a sequence. For example, states:

::= [] [] and later ::= [] ]]>

The current RBNF notation does not capture the semantics/intent of the messages; notably, when two options are mutually exclusive and at least one is mandatory. In most cases, this is noted as both options being optional. For example states:

::= [] [] [] ]]> with this example, a message that contains a response of the form

<..> ]]> (that is, a NO-PATH object followed by a path) is correct and successfully parsed. Likewise, a response with just an RP object is valid. Although the actual text within the RFC may state the intention and disambiguate the grammar, the RBNF notation can be improved to better capture the semantics, message structure and original intent. Such enhancements allow the automated validation of message elements.

Similarly, if the intent is to specific a rule such as metric-pce which includes a PCE-ID object followed by a PROC-TIME object and/or an OVERLOAD object, the syntax:

::= [] [] ]]> allows, amongst other combinations, that neither PROC-TIME nor OVERLOAD appears, which is not the intended behavior (there should be at least one metric). The alternative

::=

::= [] ::= | ]]>

or equivalently

::= (...) ::= | ]]> does not reflect that each metric-argument should appear at most once. This can be addressed verbosely:

::= ( | | ) ::= ( [] | [] )]]> Here the semantic is that we require any object of the set {PROC-TIME, OVERLOAD} to be present, and there should be at least one. Note that currently there are only a few cases where the "non-empty set" case arises.

This document does not modify the content of defined PCEP objects and TLVs. This document is not normative, the normative definition is included in the existing specs. This does not preclude integration with a future revision of such documents.

This draft does not provide any new extensions to PCEP, but it includes requirements specified by existing RFCs for illustrative purpose. The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in .

This section provides the proposed RBNF notation for the PCEP messages. Specific constructs or grammar rules that appear in several messages or deserve special considerations are described first.

::= [] ::= [] ::= [] ::= [] -- (note: named pce-list in original)]]>

A svec-tuple is a construct that associates a SVEC object with one or more constraining objects. The selected order follows the relative order of having OF and metric-list after the SVEC object, and the name svec-list has been changed since it no longer means a list of SVEC objects.

::= [] [] [] [] [] ::= []]]> Note that, again, as an example defines:

::= [] [] [] [] [] [] ]]> There are two problems, ordering and naming. So, we use the afore defined svec-tuple-list. The construct is updated to reflect the new name and to have the same relative order in the attributes that constrain a individual request

A metric-pce-id is a rule that associates a PCE identified by its PCE-ID to a list of metric arguments.

::= ( [] | [] ) ::= [] ]]>

See for the definition of specific/general and in-band/out-of-band.

::=

::= [] ::= ( | ) ::= [] ::= [] ::=

::=

::= ]]>

Attributes are used to constrain a request, or to qualify a path (defined later in this document). However, it is not straightforward to define an attributes construct, since it may change for P2P or P2MP paths, and some objects (e.g. BANDWIDTH) may appear multiple times, with different semantics: In the BANDWIDTH object can optionally appear as a path attribute or as a request constraint. In the RRO object is only used in requests "The RRO is exclusively carried within a PCReq message" for reoptimization. In such contexts, the RRO and an optional BANDWIDTH objects are bound together, in the so called rro-bw-pair construct which is also an attribute. In some contexts (stateful) paths are defined as having an optional RRO object, outside the PCEP attributes construct. In P2MP paths, multiple RRO objects may appear.

A path is defined consistently as a qualified ERO (or ERO/SERO for P2MP). Similar path constructs appear, notably, in PCEP responses, in solicited/unsolicited state reports and in update requests. The following remarks apply: The <path> construct is then defined as:

::= ( | ) [] ::= [] ::= [] ::= | [] ]]>

::= ]]>

Note that the actual parsing depends on the content (flags) of the Request Parameters (RP) object, notably expansion and P2MP. In some cases, this may be considered redundant, e.g. the presence of a PATH_KEY object and the corresponding flag. [Editor's note: from a notation perspective, we lack a way to express "if object a field x has value v then include object b, else include object c". RNBF extensions can be considered in future revisions of the PCEP protocol, e.g. defining new constructs :

) | ( ) ]]> this issue is still open.] The PCReq message contains a possibly monitored list of requests, some of which may be grouped by SVEC tuples.
::= [] [] where: ::= [] -- A request is either an expansion, a P2P request or a P2MP request ::= | | ::= ::= [] [] ::= [] -- For a P2P computation -- in RFC6006 there is a bw per tree, -- it is intended to be an optimization for an RRO list ::= (|) ::= [] ::= ]]>

::= [] -- Note: should clarify the use of SVEC tuple list where ::= [] -- An individual response may include monitoring info ::= [] [] ( | ) [] -- Note: should clarify P2MP attributes. P2MP response -- also includes endpoint-path-pair-list. TBD ::= ::= [] ]]>

The PCMonReq message is defined in for out-of-band monitoring requests. specifies that there is one mandatory object but the grammar also includes PCC-ID-REQ as mandatory. [Ed note:does it make sense to include a pce-id-list and a svec-list/request-list at the same time?]
::= [[] ] ]]>

The PCMonRep message is defined in for out-of-band monitoring responses. specifies that there is one mandatory object but the grammar also includes PCC-ID-REQ as mandatory. does not allow bundling several specific monitoring responses. A PCMonReq message causes N PCMonRep messages.
::= ]]>

::= ( | ) where ::= ::= ::= [] ::= [] ]]>

Errors can occur during PCEP handshake, or bound to one or more requests. An error during handshake is never solicited, i.e., not associated to a list of requests. A solicited error binds one or more Requests (RPs) to one or more PCEP-ERROR objects.
::= ( | ) where -- Solicited error is bound to a Request Paramters (RP) list or -- to a Stateful Request Parameters (SRP) list ::= | -- Unsolicited Error can be due to handshake or asynchronous ::= | -- Handshake Error is bound to an OPEN object ::= ::= [] ::= [] ::= []]]>

The PCRpt format is defined in . Note, however, that the end-of-sync, solicited-report and unsolicited-report are introduced for convenience, and that the RRO object is already part of the attributes construct.
::= Where: ::= [] ::= | | -- LSP flags signal end of synchronization ::= ::= ::= ]]>

As .
::= Where: ::= [] ::= ]]>

As . Note that the <path> construct is used here.
::= Where: ::= [] -- A request can be an instantiation or a deletion. SRP / LSP -- flags are used to select ::= | ) ::= ::= ]]>

This document does not define additional management considerations.

This document does not define additional security considerations.

Robert Varga Pantheon robert.varga@pantheon.sk Jonathan Harwick Metaswitch Jonathan.Hardwick@metaswitch.com Olivier Dugeon Orange olivier.dugeon@orange.com Julien Meuric Orange julien.meuric@orange.com Ina Minei Google inaminei@google.com

This work was supported in part by the PACE Support Action (http://ict-pace.net/) project under grant agreement number 619712.