Path Computation Element (PCE) Database Requirements

Looking to the different RFCs that describe the PCE architecture and in particular RFC 4655, RFC 5440, RFC 5441 and RFC 6805, the Path Computation Element (PCE) needs to acquire a set of information that is usually store in the Traffic Engineering Database (TED) in order to perform its path computation. Even if intra-domain topology acquisition is well documented and known (e.g. by listening to the IGP-TE protocol that runs inside the network), inter-domain topology information, PCE peer address, neighbor AS, existing MPLS-TE tunnels... that are necessary for the Global Concurrent Optimization, Backward Recursive Path Computation (BRPC) and the Hierarchical PCE are not documented and not completely standardized. The purpose of this memo is to inventory the required information that should be part of the PCE Database and the different mechanisms that allow an operator to populate it.

In some cases, both the path computation and the Database operations are slightly coupled: border node identification, endpoint localization, TE-LSP learning and domain sequence selection... to name a few in which an IGP-based TED may not be sufficient. It is also important to differentiate several environments with different requirements, especially for the multi-domain problem. The PCE is scoped for any kind of network, from transmission networks (TDM/WDM) with a rather limited number of domains, few interconnections, and few confidentiality issues; transmission networks with a large number of domains; MPLS networks with several administrative domains; and big IP/MPLS networks with a large number of domains with peering agreements. For each of them, a different solution for the multi-domain path computation may apply. A solution may not be scalable for one, but perfectly suitable for another. Up to now, PCE WG has based its work and standard on the assumption and hypothesis that the TED contains all pertinent information suitable for the PCE to compute an optimal TE-LSP placement, over one or several domains a PCE has visibility on or over a set of PCE-capable domains (e.g. using BRPC procedure). We could identify several major sources of information for the TED: The intra-domain routing protocol like OSPF-TE or IS-IS-TE (including extensions for inter-domain links), The inter-domain routing protocol, i.e. BGP, TED synchronization protocols, e.g., BGP-LS, Through manual and or management configuration. If the first source gives a precise and synchronize view of the controlled network, i/eBGP typically just provides network reachability with only one AS path (unless using recent add path option). Now, the TE information traditionally flooded by the IGPs can also be communicated through a BGP sessions, as described in "North bound distribution of Link-State and TE Information using BGP". Nevertheless, to optimize inter-domain path computation, route diversity and a minimum set of Traffic Engineering information about the remote domains could be helpful. Despite that it is possible to re-announce TE-LSP in the IGP-TE, the PCE needs also to have a precise knowledge of previous TE-LSP, not only for its stateful version [PCEP Extensions for Stateful PCE], but also when performing a global concurrent optimization RFC5557 of the previous TE-LSPs place on a given domain. The last source of information, mainly static, information can be the management plane, e.g. using SNMP, Netconf, CLI... So, it is necessary to classify the source of information by their frequency of update: static or dynamic, e.g. a domain ID is unlikely to change, while unreserved bandwidth of a link may be continuously changing. Finally, all sources of information are pertinent and must be take into account to fulfil the PCE database at large. In this document, PCE Data Base (namely PCE-DB in the rest of the document) is used not only to refer to the usual notion of Traffic Engineering Database information, but also encompasses all relevant information. E.g., the phrase also refers to the list of TE-LSPs running in the domain, sometimes referred as LSP-DB in other documents. Note that this PCE-DB may be implemented over multiple independant DBs.

ABR: Area Border Routers. Routers used to connect two IGP areas (areas in OSPF or levels in IS-IS). ASBR: Autonomous System Border Router. Router used to connect together ASes of the same or different service providers via one or more inter-AS links. AS: Autonomous System Boundary Node (BN): a boundary node is either an ABR in the context of inter-area Traffic Engineering or an ASBR in the context of inter-AS Traffic Engineering. Domain: an Autonomous System or IGP Area Entry BN of domain(n): a BN connecting domain(n-1) to domain(n) along a determined sequence of domains. Exit BN of domain(n): a BN connecting domain(n) to domain(n+1) along a determined sequence of domains. Inter-area TE LSP: A TE LSP that crosses an IGP area boundary. Inter-AS TE LSP: A TE LSP that crosses an AS boundary. IGP-TE: Interior Gateway Protocol with Traffic Engineering support. Both OSPF-TE and IS-IS-TE are identified in this category. PCE: Path Computation Element. An entity (component, application, or network node) that is capable of computing a network path or route based on a network graph and applying computational constraints. PCE(i) is a PCE with the scope of domain(i). PCE-DB: Path Computation Element Data Base TED: Traffic Engineering Database.

This section made a first inventory of the main requirements of the PCE Data Base in term of information that the database should contains.

This section describes the Intra-domain information that are suitable for the PCE Database including both MPLS and GMPLS.

A PCE is allowed to compute paths in one or several domains. Such PCE MUST be aware of the precise details of the network topology (or topologies) in order to compute optimal TE-LSP placements. The information needed in this case includes: List of Internal Nodes identified by a reachable address: All nodes of the networks with a particular mention for border node (see next section), List of Internal Links that rely nodes (both internal and border nodes), Traffic Engineering information of the different links i.e. RFC 3630 and RFC 5305(with e.g. recent metric extensions proposal OSPF Traffic Engineering (TE) Metric Extensions) Traffic Engineering information of the nodes. The information above mentioned is usually exchanged using the IGP-TE protocol (OSPF-TE or IS-IS-TE).

When dealing with a (G)MPLS network, the PCE MUST be aware of the complementary information: Traffic Engineering information with GMPLS extensions of the different links i.e. RFC 4203 and RFC 5307, To be completed latter

A PCE can also be allowed to take part to inter-domain path computation (e.g in per-domain path computation, BRPC or H-PCE relationship). Some inter-domain information is mandatory when operator intend to use the PCE to compute Inter-AS TE LSP path that cross domain boundary. For that purpose, the PCE-DB SHOULD contains all information that allow the PCE to determine the optimal inter-domain path for the TE-LSP computation, which includes: Border Nodes (BNs) of the domain. A distinction could be made between ALL domains and Neighbor domains only. In the document, we consider ALL domains to be sure that the PCE has the complete visibility of the path diversity. Links between BN, i.e. links between BN (n) to BN (n+1), including Traffic Engineering information, Traffic Engineering performance between BN (n) to give performance indication on remote domain n (See section 3.2 on PCE-DB model for the inter-domain part) PCE (i) peer address associated with the domain number and identity of the remote domain (i), RFC 5316 for IS-IS and RFC 5392 for OSPF help to provide required PCE-DB information in the case of inter-domain. PCE-DB can also contain information about virtual links and abstract information.

For stateful operation and Global Concurrent Optimization, the PCE-DB should also contain information on TE-LSPs already enforce in the controlled domain. If some TE-LSP tunnels could be re-announce in the IGP-TE, the PCE could not learn from the IGP-TE all details of all TE LSPs: if TE information is known, detail of the ERO is lost as well as initial QoS parameters. The following information will be useful for the PCE-DB to describe the TE-LSP: Explicit Route Object (ERO), End-points objects, Initial and actual Metric objects, including extend metrics such as delay, jitter, packet loss, Recent PCEP Extensions for Stateful PCE provide new PCEP message to convey these kind of information. However, this capacity could be used disregarding the behavior (stateless or stateful) of the PCE. Indeed, if it is mandatory for stateful PCE, these information are of great importance then performing Global Concurrent Optimization, even with a stateless PCE.

This part of the TED contains all others information, and in particular the PCE policy, pertinent for the PCE to compute TE LSP path but that are provided through the management system.

This section inventory the database model(s) to store pertinent information regarding the different source of information.

For intra-domain, there is no need to specify a particular model or schema for the PCE-DB. Indeed, the model is directly based on the IGP-TE. Of course there is a difference between IS-IS and OSPF, but TE Link state are more of less similar in term of conveyed information and database description. No particular requirements are necessary as this stage.

To be provided later.

Contrary to intra-domain where the PCE known the exact details of the underlying network, it is not possible to achieve a similar detail level for the inter-domain. And not only for scalability reasons, but mostly for confidentiality of the networks. This memo propose a basic schema that allows PCE to known sufficient details about the remote domain while keeping confidential the internal information. For this purpose, we propose to describe a domain as a "Grey-Box" with inputs and outputs that correspond to the Border Nodes (BNs). Then Grey-Boxes are interconnected through inter-domain links between the BNs. Then, suitable performance indicators are given to cross the Grey-Boxes from an input BN to and output BN. Figure below gives as example of such model.

Domain C is reachable from domain A through domain B or domain D. For a PCE, with such model, it becomes easy to compute a constraint shortest path by combining the resources availability on Inter-Domain links and cost to cross the different domain. For example, with these figures (note that we take only one measured to illustrate the purpose and that multiple constraints are used in reality): Crossing B cost 100, D cost 50, C cost 75, A cost 50 Inter-domain links costs: A to B = 10, B to C = 20, A to D = 10, D to C = 50 PCE A could not choose between B or D as the Inter-domain link costs are equal. With the proposed model, it could compute that going through B cost 130 (= 10 + 100 20) and through D cost 110 (= 10 + 50 + 50) and choose D path even if the last Inter-domain links is costly. Today, when trying to compute an inter-domain TE LSP, the PCE may failed in its computation and used crank back facilities to find a suitable path. With such inter-domain information, a PCE could look into the different inter-domain path (as the sum of inter-domain links and Grey-Box crossing performances) and select the most suitable one regarding the PCReq avoiding crank back and achieve possibly, better results as it explore all possible inter-domain paths. If the inter-domain links between BN that connect the Grey-Boxes description are covered (see section 2.2), it is not the case for the internal links between BNs inside the Grey-Box.

This section aims to provide best current practices when mechanisms are well-known and some hints when standard solutions exist to populate the PCE TED, and so give directions to extend them. In particular, we aim at providing input on whether the TED gets the information from the routing protocol and how it gets it, which specific routing protocols are suited, whether it gets it from an NMS, at what frequency the TED is updated... and if it needs extra information.

As the TED mainly contains the intra-domain topology graph, it is RECOMMENDED to link the PCE with the underlying IGP-TE (OSPF-TE or IS-IS-TE routing protocol). By adding the PCE into the IGP-TE routing intra-domain, it is possible to listen to the routing protocol and then acquired the complete topology graph as well as let the PCE announce itself (see RFC5088 and RFC5089). In addition, the TED will synchronize as fast as the routing protocol converges like any router in the domain. Best current practices are also of interest when a PCE compute path that spawn to several area / region. In that case, the PCE must be aware of the topology details of each area / region. Note that linking the PCE with the underlying IGP-TE may also be accomplished through receiving BGP-LS updates as described in "North bound distribution of Link-State and TE Information using BGP". Although joining the IGP is good enough, BGP-LS is not precluded for use intra-domain and can be a nice way to have a uniform mechanism to acquire the TED e.g. from a Route Reflector that also listen to the IGP. In addition, management tools may be used to complement the topology graph provided by the routing protocol.

To be provided later.

If for inter-area aspect of the inter-domain, actual IGP protocol provide in general the aforementioned information without any particular extension, this is not the case for the inter-as scenario and sometimes an issue for inter-area. First of all, RFC 5316 or RFC 5392 MUST be activated in the IGP-TE (respectively in IS-IS-TE or OSPF-TE) in order to advertise TE information on the inter-domain links. This gives the advantage for the PCE to determine what could be feasible, during path computation, on the peering links. In MPLS, AS path and network reachability are obtained from BGP and routing tables. In addition, domain or sequence path could be specified in the PCE Request. However, when inter-domain path is not known or could not retrieve from an external entities, it could be of interest for a PCE to have the possibility to compute the inter-domain path prior to the intra-domain part. Again, the PCE needs corresponding information in its PCE-DB. But, it is not straightforward to collect route diversity or TE information (i.e. bandwidth, transit delay, packet loss ratio, jitter ...) on a remote domain. Of course, for confidentiality and scalability issues, the PCE MUST NOT learn all details of the remote TED, it just needs an abstract view like proposed in "Problem Statement and Architecture for Information Exchange Between Interconnected Traffic Engineered Networks". Right now, we have identified several methods, which have been tested to fill in the PCE-DB with this kind of information: Use of the management plane; Use of the "North bound distribution of Link-State and TE Information using BGP" proposal to exchange TE information about the remote domain; Use of PCNtf message to convey, inside vendor attribute (but in an extended way), TE information of remote domain between PCE As well as some potential alternative mechanisms that would need more standardization effort: A Hierarchical TE that could help to advertise, at the AS level, TE information on an abstract view of the remote AS topology; A PCEP extension to convey such TE information to the remote PCE.

The force of PCE is to be aware of the complete topology of the underlying network where it is connected. With such knowledge, it could place efficiently the tunnel even if it not follows the route computed by the IGP routing protocol. Same principles apply also for the inter-domain. But, in the Internet today, BGP summarize the route and the PCE should not be aware of the route diversity. In particular, it could not choose another AS path as the one selected and announced by BGP. A way to bypass this restriction is to specify the AS-path in the PCE Request IRO. In all other cases, the PCE will not be sufficiently aware of the route diversity and can not select the optimal AS path when computing an inter-domain LSP. To avoid this and allow PCE to know route diversity to reach a given remote domain, the inter-domain information must be propagated between all PCEs without aggregation or summarization. In summary, PCEs need to synchronize part of their Database i.e. the inter-domain part. Disregarding the protocol, two different solutions emerged to exchange inter-domain information: Direct Distribution: Exchange TE information using BGP is part of this case. In this scenario, it is necessary to establish a BGP session between the different domains (whatever the platform used, a dedicated router, a PCE, another server ...). In the hierarchal PCE scenario, operators that provide child PCE, agree to establish a relation with remote domain that provides the parent PCE. But, in BRPC, or in Hierarchical PCE where almost operators provide a parent PCE, BGP session must be establish between networks that have not necessary direct adjacency. However, operators should not agree to accept relation from other's not directly attached to their network. In addition, this scenario could conduct to establish a full mesh of BGP session between PCE which could lead into some scalability problems. Flooding Distribution: In this case, the inter-domain information are flood between all PCE so that each PCE is aware about all remote domain capabilities. This meets the requirement but doesn't provide the flexibility of BGP in term of filtering. Indeed, BGP allows through configuration to decide which information are announced and to whom. As a per session relation, a given operator is not oblige to announce the same capabilities to its remote domain. With flooding distribution, where everybody redistribute what it has learned without modify it, it is not possible to specialize announcement based on remote domain. So, the solution must provide the possibility to filter what is announce per remote domain without authorized the summarization or aggregation while keeping a distributed relation between domains. In addition, a domain is responsible about the Grey-Box announcement and the advertisement information must not be modified by intermediate PCE.

Up to know, the PCE could learn the tunnel already enforce in the controlled domain through dedicated NMS system. Recent works on state full extensions for PCEP propose to add new messages in order to collect information on TE-LSPs from the PCCs.

Most of the time, static information, including PCE Policy, are provided through the management system of the operator or by means of static configuration (e.g. command line option, configuration file ...), but some could be automatically discovered. In particular, in intra-domain, PCCs and PCEs can discover automatically reachable PCEs (as well as computation domains) through the deployment of RFC 5088, for OSPF-controlled networks, and RFC 5089 for IS-IS controlled networks. However, for the inter-PCE discovery at the inter-AS level, no mechanism has been standardized (unless ASes are owned by the same ISP).

Even if acquired TE information is solved, it remains two major problems from an operational point of view. First of all, the PCE-DB MUST be synchronised with the underlying network topology. This synchronisation is not only mandatory for the efficiency of the answer of the PCE, but more to handle the graceful restart step of the PCE as well as crash situation. Indeed, for divert reasons (maintenance, scheduled operation, failure ...), when the PCE start or restart, it MUST acquired the information of the PCE-DB and then maintain it synchronised to the underlying network. For the stateful version of the PCE, this synchronisation is mandatory as TE-LSP tunnel could be setup manually or by the management plane independently from the PCE. But, the PCE MUST be aware of them as well as when the PCE restart is MUST be aware of TE-LSP it previously setup. The second point come from the distributed nature of the TED information located in the underlying network. Indeed, TE information are not located in one place, but distributed amongst all the router of the network. Each router manage its links, and, consequently, the TE information attached to these links. Thus, modifying a TE information on large scale network could become quickly a nightmare for operational without any tools to help them. For that purpose, a TE Netconf model like proposed in "A YANG Data Model for Network Topologies" is mandatory from an operation point of view to allow automatic tools easily configure the TE parameters of a network on the routers.

This document makes no request of IANA for the moment. Note to RFC Editor: this section may be removed on publication as an RFC.

Acquisition of information for the PCE TED is of course sensible from a security point of view, especially when acquiring information from others AS. This section aims at providing best practices to prevent some security threat when the PCE try to acquire TED information.

Same security considerations must be applied to the PCE when it is connected to an IGP-TE protocol as the routing protocol itself. Best practices observed and deployed by operators must also be taken into account when installing some PCEs. Indeed, even when deployed as a standalone server, PCEs must be considered as a typical router from the IGP-TE perspective. As a result, beyond OSPF or IS-IS themselves, the usual security rules must be applied, e.g. login/passwd, authentication/digest... to protect the connectivity.

Inter-domain relation and so information exchange are subject to high potential hijack and so need attention from the security point of view. To avoid disclosing or expose confidential information that two operators would exchange to fill in the TEDs of their respective PCEs, the relation SHOULD be protected by standard cryptography mechanism. E.g. using IPsec tunnel is RECOMMENDED to protect the connectivity between PCEs and the TED exchanges.

All operational information like PCE peer addresses are generally added manually to the TED and so do not need any particular protection nor subject to security. But, as this basic information is needed to connected the PCEs to their peers, it could potentially be associated to sensitive parameters like login and password. So, standard Best Practices are RECOMMENDED to avoid basic security exposition.

The authors want to thanks PCE's WG members and in particular Daniel King for their inputs of this subject.