RFC 9522: Overview and Principles of Internet Traffic Engineering

1.1. What is Internet Traffic Engineering?

One of the most significant functions performed in the Internet is the routing and forwarding of traffic from ingress nodes to egress nodes. Therefore, one of the most distinctive functions performed by Internet traffic engineering is the control and optimization of these routing and forwarding functions, to steer traffic through the network.¶

Internet traffic engineering is defined as that aspect of Internet network engineering dealing with the issues of performance evaluation and performance optimization of operational IP networks. Traffic engineering encompasses the application of technology and scientific principles to the measurement, characterization, modeling, and control of Internet traffic [RFC2702] [AWD2].¶

It is the performance of the network as seen by end users of network services that is paramount. The characteristics visible to end users are the emergent properties of the network, which are the characteristics of the network when viewed as a whole. A central goal of the service provider, therefore, is to enhance the emergent properties of the network while taking economic considerations into account. This is accomplished by addressing traffic-oriented performance requirements while utilizing network resources without excessive waste and in a reliable way. Traffic-oriented performance measures include delay, delay variation, packet loss, and throughput.¶

Internet TE responds to network events (such as link or node failures, reported or predicted network congestion, planned maintenance, service degradation, planned changes in the traffic matrix, etc.). Aspects of capacity management respond at intervals ranging from days to years. Routing control functions operate at intervals ranging from milliseconds to days. Packet-level processing functions operate at very fine levels of temporal resolution (up to milliseconds) while reacting to statistical measures of the real-time behavior of traffic.¶

Thus, the optimization aspects of TE can be viewed from a control perspective and can be both proactive and reactive. In the proactive case, the TE control system takes preventive action to protect against predicted unfavorable future network states, for example, by engineering backup paths. It may also take action that will lead to a more desirable future network state. In the reactive case, the control system responds to correct issues and adapt to network events, such as routing after failure.¶

Another important objective of Internet TE is to facilitate reliable network operations [RFC2702]. Reliable network operations can be facilitated by providing mechanisms that enhance network integrity and by embracing policies emphasizing network survivability. This reduces the vulnerability of services to outages arising from errors, faults, and failures occurring within the network infrastructure.¶

The optimization aspects of TE can be achieved through capacity management and traffic management. In this document, capacity management includes capacity planning, routing control, and resource management. Network resources of particular interest include link bandwidth, buffer space, and computational resources. In this document, traffic management includes:¶

One major challenge of Internet TE is the realization of automated control capabilities that adapt quickly and cost-effectively to significant changes in network state, while still maintaining stability of the network. Performance evaluation can assess the effectiveness of TE methods, and the results of this evaluation can be used to identify existing problems, guide network reoptimization, and aid in the prediction of potential future problems. However, this process can also be time-consuming and may not be suitable to act on short-lived changes in the network.¶

Performance evaluation can be achieved in many different ways. The most notable techniques include analytic methods, simulation, and empirical methods based on measurements.¶

Traffic engineering comes in two flavors:¶

In the latter case, any deviation from the optimum distribution (e.g., caused by a fiber cut) is reverted upon repair without further optimization. However, this form of TE relies upon the notion that the planned state of the network is optimal. Hence, there are two levels of TE in such a mode:¶

As a general rule, TE concepts and mechanisms must be sufficiently specific and well-defined to address known requirements but simultaneously flexible and extensible to accommodate unforeseen future demands (see Section 6.1).¶

1.2. Components of Traffic Engineering

As mentioned in Section 1.1, Internet traffic engineering provides performance optimization of IP networks while utilizing network resources economically and reliably. Such optimization is supported at the control/controller level and within the data/forwarding plane.¶

The key elements required in any TE solution are as follows:¶

Some TE solutions rely on these elements to a lesser or greater extent. Debate remains about whether a solution can truly be called "TE" if it does not include all of these elements. For the sake of this document, we assert that all TE solutions must include some aspects of all of these elements. Other solutions can be classed as "partial TE" and also fall in scope of this document.¶

Policy allows for the selection of paths (including next hops) based on information beyond basic reachability. Early definitions of routing policy, e.g., [RFC1102] and [RFC1104], discuss routing policy being applied to restrict access to network resources at an aggregate level. BGP is an example of a commonly used mechanism for applying such policies; see [RFC4271] and [RFC8955]. In the TE context, policy decisions are made within the control plane or by controllers in the management plane and govern the selection of paths. Examples can be found in [RFC4655] and [RFC5394]. TE solutions may cover the mechanisms to distribute and/or enforce policies, but definition of specific policies is left to the network operator.¶

Path steering is the ability to forward packets using more information than just knowledge of the next hop. Examples of path steering include IPv4 source routes [RFC0791], RSVP-TE explicit routes [RFC3209], Segment Routing (SR) [RFC8402], and Service Function Chaining [RFC7665]. Path steering for TE can be supported via control plane protocols, by encoding in the data plane headers, or by a combination of the two. This includes when control is provided by a controller using a network-facing control protocol.¶

Resource management provides resource-aware control and forwarding. Examples of resources are bandwidth, buffers, and queues, all of which can be managed to control loss and latency.¶

Resource reservation is the control aspect of resource management. It provides for domain-wide consensus about which network resources are used by a particular flow. This determination may be made at a very coarse or very fine level. Note that this consensus exists at the network control or controller level but not within the data plane. It may be composed purely of accounting/bookkeeping, but it typically includes an ability to admit, reject, or reclassify a flow based on policy. Such accounting can be done based on any combination of a static understanding of resource requirements and the use of dynamic mechanisms to collect requirements (e.g., via RSVP-TE [RFC3209]) and resource availability (e.g., via OSPF extensions for GMPLS [RFC4203]).¶

Resource allocation is the data plane aspect of resource management. It provides for the allocation of specific node and link resources to specific flows. Example resources include buffers, policing, and rate-shaping mechanisms that are typically supported via queuing. Resource allocation also includes the matching of a flow (i.e., flow classification) to a particular set of allocated resources. The method of flow classification and granularity of resource management is technology-specific. Examples include Diffserv with dropping and remarking [RFC4594], MPLS-TE [RFC3209], GMPLS-based Label Switched Paths (LSPs) [RFC3945], as well as controller-based solutions [RFC8453]. This level of resource control, while optional, is important in networks that wish to support network congestion management policies to control or regulate the offered traffic to deliver different levels of service and alleviate network congestion problems. It is also important in networks that wish to control the latency experienced by specific traffic flows.¶

1.3. Scope

The scope of this document is intra-domain TE because this is the practical level of TE technology that exists in the Internet at the time of writing. That is, this document describes TE within a given AS in the Internet. This document discusses concepts pertaining to intra-domain traffic control, including such issues as routing control, micro and macro resource allocation, and control coordination problems that arise consequently.¶

This document describes and characterizes techniques already in use or in advanced development for Internet TE. The way these techniques fit together is discussed and scenarios in which they are useful are identified.¶

Although the emphasis in this document is on intra-domain traffic engineering, an overview of the high-level considerations pertaining to inter-domain TE is provided in Section 7. Inter-domain Internet TE is crucial to the performance enhancement of the world-wide Internet infrastructure.¶

Whenever possible, relevant requirements from existing IETF documents and other sources are incorporated by reference.¶

1.4. Terminology

This section provides terminology that is useful for Internet TE. The definitions presented apply to this document. These terms may have other meanings elsewhere.¶

Busy hour:

A one-hour period within a specified interval of time (typically 24 hours) in which the traffic load in a network or sub-network is greatest.¶

Congestion:

A state of a network resource in which the traffic incident on the resource exceeds its output capacity over an interval of time. A small amount of congestion may be beneficial to ensure that network resources are run at full capacity, and this may be particularly true at the network edge where it is desirable to ensure that user traffic is served as much as possible. Within the network, if congestion is allowed to build (such as when input traffic exceeds output traffic in a sustained way), it will have a negative effect on user traffic.¶

Congestion avoidance:

An approach to congestion management that attempts to obviate the occurrence of congestion. It is chiefly relevant to network congestion, although it may form a part of demand-side congestion management.¶

Congestion response:

An approach to congestion management that attempts to remedy congestion problems that have already occurred.¶

Constraint-based routing:

A class of routing protocols that takes specified traffic attributes, network constraints, and policy constraints into account when making routing decisions. Constraint-based routing is applicable to traffic aggregates as well as flows. It is a generalization of QoS-based routing.¶

Demand-side congestion management:

A congestion management scheme that addresses congestion problems by regulating or conditioning the offered load.¶

Effective bandwidth:

The minimum amount of bandwidth that can be assigned to a flow or traffic aggregate in order to deliver "acceptable service quality" to the flow or traffic aggregate. See [KELLY] for a more mathematical definition.¶

Egress node:

The device (router) at which traffic leaves a network toward a destination (host, server, etc.) or to another network.¶

End-to-end:

This term is context-dependent and often applies to the life of a traffic flow from original source to final destination. In contrast, edge-to-edge is often used to describe the traffic flow from the entry of a domain or network to the exit of that domain or network. However, in some contexts (for example, where there is a service interface between a network and the client of that network or where a path traverses multiple domains under the control of a single process), end-to-end is used to refer to the full operation of the service that may be composed of concatenated edge-to-edge operations. Thus, in the context of TE, the term "end-to-end" may refer to the full TE path but not to the complete path of the traffic from source application to ultimate destination.¶

Hotspot:

A network element or subsystem that is in a considerably higher state of congestion than others.¶

Ingress node:

The device (router) at which traffic enters a network from a source (host) or from another network.¶

Metric:

A parameter defined in terms of standard units of measurement.¶

Measurement methodology:

A repeatable measurement technique used to derive one or more metrics of interest.¶

Network congestion:

Congestion within the network at a specific node or a specific link that is sufficiently extreme that it results in unacceptable queuing delay or packet loss. Network congestion can negatively impact end-to-end or edge-to-edge traffic flows, so TE schemes may be deployed to balance traffic in the network and deliver congestion avoidance.¶

Network survivability:

The capability to provide a prescribed level of QoS for existing services after a given number of failures occur within the network.¶

Offered load:

Offered load is also sometimes called "offered traffic load". It is a measure of the amount of traffic being presented to be carried across a network compared to the capacity of the network to carry it. This term derives from queuing theory, and an offered load of 1 indicates that the network can carry, but only just manage to carry, all of the traffic presented to it.¶

Offline traffic engineering:

A traffic engineering system that exists outside of the network.¶

Online traffic engineering:

A traffic-engineering system that exists within the network, typically implemented on or as adjuncts to operational network elements.¶

Performance measures:

Metrics that provide quantitative or qualitative measures of the performance of systems or subsystems of interest.¶

Performance metric:

A performance parameter defined in terms of standard units of measurement.¶

Provisioning:

The process of assigning or configuring network resources to meet certain requests.¶

Quality of Service (QoS):

QoS [RFC3198] refers to the mechanisms used within a network to achieve specific goals for the delivery of traffic for a particular service according to the parameters specified in a Service Level Agreement. "Quality" is characterized by service availability, delay, jitter, throughput, and packet loss ratio. At a network resource level, "Quality of Service" refers to a set of capabilities that allow a service provider to prioritize traffic, control bandwidth, and network latency.¶

QoS routing:

Class of routing systems that selects paths to be used by a flow based on the QoS requirements of the flow.¶

Service Level Agreement (SLA):

A contract between a provider and a customer that guarantees specific levels of performance and reliability at a certain cost.¶

Service Level Objective (SLO):

A key element of an SLA between a provider and a customer. SLOs are agreed upon as a means of measuring the performance of the service provider and are outlined as a way of avoiding disputes between the two parties based on misunderstanding.¶

Stability:

An operational state in which a network does not oscillate in a disruptive manner from one mode to another mode.¶

Supply-side congestion management:

A congestion management scheme that provisions additional network resources to address existing and/or anticipated congestion problems.¶

Traffic characteristic:

A description of the temporal behavior or a description of the attributes of a given traffic flow or traffic aggregate.¶

Traffic-engineering system:

A collection of objects, mechanisms, and protocols that are used together to accomplish traffic-engineering objectives.¶

Traffic flow:

A stream of packets between two endpoints that can be characterized in a certain way. A common classification for a traffic flow selects packets with the five-tuple of source and destination addresses, source and destination ports, and protocol ID. Flows may be very small and transient, ranging to very large. The TE techniques described in this document are likely to be more effective when applied to large flows. Traffic flows may be aggregated and treated as a single unit in some forms of TE, making it possible to apply TE to the smaller flows that comprise the aggregate.¶

Traffic mapping:

Traffic mapping is the assignment of traffic workload onto (pre-established) paths to meet certain requirements.¶

Traffic matrix:

A representation of the traffic demand between a set of origin and destination abstract nodes. An abstract node can consist of one or more network elements.¶

Traffic monitoring:

The process of observing traffic characteristics at a given point in a network and collecting the traffic information for analysis and further action.¶

Traffic trunk:

An aggregation of traffic flows belonging to the same class that are forwarded through a common path. A traffic trunk may be characterized by an ingress and egress node and a set of attributes that determine its behavioral characteristics and requirements from the network.¶

Workload:

Workload is also sometimes called "traffic workload". It is an evaluation of the amount of work that must be done in a network in order to facilitate the traffic demand. Colloquially, it is the answer to, "How busy is the network?"¶

2.1. Context of Internet Traffic Engineering

The context of Internet traffic engineering includes the following sub-contexts:¶

The context of Internet TE and the different problem scenarios are discussed in the following subsections.¶

2.2. Network Domain Context

IP networks range in size from small clusters of routers situated within a given location to thousands of interconnected routers, switches, and other components distributed all over the world.¶

At the most basic level of abstraction, an IP network can be represented as a distributed dynamic system consisting of:¶

The network elements and resources may have specific characteristics restricting the manner in which the traffic demand is handled. Additionally, network resources may be equipped with traffic control mechanisms managing the way in which the demand is serviced. Traffic control mechanisms may be used to:¶

A configuration management and provisioning system may allow the settings of the traffic control mechanisms to be manipulated by external or internal entities in order to exercise control over the way in which the network elements respond to internal and external stimuli.¶

The details of how the network carries packets are specified in the policies of the network administrators and are installed through network configuration management and policy-based provisioning systems. Generally, the types of service provided by the network also depend upon the technology and characteristics of the network elements and protocols, the prevailing service and utility models, and the ability of the network administrators to translate policies into network configurations.¶

Internet networks have two significant characteristics:¶

The dynamic characteristics of IP and IP/MPLS networks can be attributed in part to fluctuations in demand, the interaction between various network protocols and processes, the rapid evolution of the infrastructure that demands the constant inclusion of new technologies and new network elements, and the transient and persistent faults that occur within the system.¶

Packets contend for the use of network resources as they are conveyed through the network. A network resource is considered to be congested if, for an interval of time, the arrival rate of packets exceeds the output capacity of the resource. Network congestion may result in some of the arriving packets being delayed or even dropped.¶

Network congestion increases transit delay and delay variation, may lead to packet loss, and reduces the predictability of network services. Clearly, while congestion may be a useful tool at ingress edge nodes, network congestion is highly undesirable. Combating network congestion at a reasonable cost is a major objective of Internet TE, although it may need to be traded with other objectives to keep the costs reasonable.¶

Efficient sharing of network resources by multiple traffic flows is a basic operational premise for the Internet. A fundamental challenge in network operation is to increase resource utilization while minimizing the possibility of congestion.¶

The Internet has to function in the presence of different classes of traffic with different service requirements. This requirement is clarified in the architecture for Differentiated Services (Diffserv) [RFC2475]. That document describes how packets can be grouped into behavior aggregates such that each aggregate has a common set of behavioral characteristics or a common set of delivery requirements. Delivery requirements of a specific set of packets may be specified explicitly or implicitly. Two of the most important traffic delivery requirements are:¶

2.3. Problem Context

There are several problems associated with operating a network like those described in the previous section. This section analyzes the problem context in relation to TE. The identification, abstraction, representation, and measurement of network features relevant to TE are significant issues.¶

A particular challenge is to formulate the problems that traffic engineering attempts to solve. For example:¶

Another class of problems is how to measure and estimate relevant network state parameters. Effective TE relies on a good estimate of the offered traffic load as well as a view of the underlying topology and associated resource constraints. Offline planning requires a full view of the topology of the network or partial network that is being planned.¶

Still another class of problem is how to characterize the state of the network and how to evaluate its performance. The performance evaluation problem is two-fold: one aspect relates to the evaluation of the system-level performance of the network, and the other aspect relates to the evaluation of resource-level performance, which restricts attention to the performance analysis of individual network resources.¶

In this document, we refer to the system-level characteristics of the network as the "macro-states" and the resource-level characteristics as the "micro-states." The system-level characteristics are also known as the emergent properties of the network. Correspondingly, we refer to the TE schemes dealing with network performance optimization at the systems level as "macro-TE" and the schemes that optimize at the individual resource level as "micro-TE." Under certain circumstances, the system-level performance can be derived from the resource-level performance using appropriate rules of composition, depending upon the particular performance measures of interest.¶

Another fundamental class of problem concerns how to effectively optimize network performance. Performance optimization may entail translating solutions for specific TE problems into network configurations. Optimization may also entail some degree of resource management control, routing control, and capacity augmentation.¶

2.4. Solution Context

The solution context for Internet TE involves analysis, evaluation of alternatives, and choice between alternative courses of action. Generally, the solution context is based on making inferences about the current or future state of the network and making decisions that may involve a preference between alternative sets of action. More specifically, the solution context demands reasonable estimates of traffic workload, characterization of network state, derivation of solutions that may be implicitly or explicitly formulated, and possibly instantiation of a set of control actions. Control actions may involve the manipulation of parameters associated with routing, control over tactical capacity acquisition, and control over the traffic management functions.¶

The following list of instruments may be applicable to the solution context of Internet TE:¶

Determining traffic characteristics through measurement or estimation is very useful within the realm of the TE solution space. Traffic estimates can be derived from customer subscription information, traffic projections, traffic models, and actual measurements. The measurements may be performed at different levels, e.g., at the traffic-aggregate level or at the flow level. Measurements at the flow level or on small traffic aggregates may be performed at edge nodes, when traffic enters and leaves the network. Measurements for large traffic aggregates may be performed within the core of the network.¶

To conduct performance studies and to support planning of existing and future networks, a routing analysis may be performed to determine the paths the routing protocols will choose for various traffic demands and to ascertain the utilization of network resources as traffic is routed through the network. Routing analysis captures the selection of paths through the network, the assignment of traffic across multiple feasible routes, and the multiplexing of IP traffic over traffic trunks (if such constructs exist) and over the underlying network infrastructure. A model of network topology is necessary to perform routing analysis. A network topology model may be extracted from:¶

Topology information may also be derived from servers that monitor network state and from servers that perform provisioning functions.¶

Routing in operational IP networks can be administratively controlled at various levels of abstraction, including the manipulation of BGP attributes and IGP metrics. For path-oriented technologies such as MPLS, routing can be further controlled by the manipulation of relevant TE parameters, resource parameters, and administrative policy constraints. Within the context of MPLS, the path of an explicitly routed LSP can be computed and established in various ways, including:¶

2.5. Implementation and Operational Context

The operational context of Internet TE is characterized by constant changes that occur at multiple levels of abstraction. The implementation context demands effective planning, organization, and execution. The planning aspects may involve determining prior sets of actions to achieve desired objectives. Organizing involves arranging and assigning responsibility to the various components of the TE system and coordinating the activities to accomplish the desired TE objectives. Execution involves measuring and applying corrective or perfective actions to attain and maintain desired TE goals.¶

3.1. Components of the Traffic-Engineering Process Model

The key components of the traffic-engineering process model are as follows:¶

4.1. Time-Dependent versus State-Dependent versus Event-Dependent

Traffic-engineering methodologies can be classified as time-dependent, state-dependent, or event-dependent. All TE schemes are considered to be dynamic in this document. Static TE implies that no TE methodology or algorithm is being applied -- it is a feature of network planning but lacks the reactive and flexible nature of TE.¶

In time-dependent TE, historical information based on periodic variations in traffic (such as time of day) is used to pre-program routing and other TE control mechanisms. Additionally, customer subscription or traffic projection may be used. Pre-programmed routing plans typically change on a relatively long timescale (e.g., daily). Time-dependent algorithms do not attempt to adapt to short-term variations in traffic or changing network conditions. An example of a time-dependent algorithm is a centralized optimizer where the input to the system is a traffic matrix and multi-class QoS requirements as described in [MR99]. Another example of such a methodology is the application of data mining to Internet traffic [AJ19], which enables the use of various machine learning algorithms to identify patterns within historically collected datasets about Internet traffic and to extract information in order to guide decision-making and improve efficiency and productivity of operational processes.¶

State-dependent TE adapts the routing plans based on the current state of the network, which provides additional information on variations in actual traffic (i.e., perturbations from regular variations) that could not be predicted using historical information. Constraint-based routing is an example of state-dependent TE operating in a relatively long timescale. An example of operating in a relatively short timescale is a load-balancing algorithm described in [MATE]. The state of the network can be based on parameters flooded by the routers. Another approach is for a particular router performing adaptive TE to send probe packets along a path to gather the state of that path. [RFC6374] defines protocol extensions to collect performance measurements from MPLS networks. Another approach is for a management system to gather the relevant information directly from network elements using telemetry data collection publication/subscription techniques [RFC7923]. Timely gathering and distribution of state information is critical for adaptive TE. While time-dependent algorithms are suitable for predictable traffic variations, state-dependent algorithms may be needed to increase network efficiency and to provide resilience to adapt to changes in network state.¶

Event-dependent TE methods can also be used for TE path selection. Event-dependent TE methods are distinct from time-dependent and state-dependent TE methods in the manner in which paths are selected. These algorithms are adaptive and distributed in nature, and they typically use learning models to find good paths for TE in a network. While state-dependent TE models typically use available-link-bandwidth (ALB) flooding [E.360.1] for TE path selection, event-dependent TE methods do not require ALB flooding. Rather, event-dependent TE methods typically search out capacity by learning models, as in the success-to-the-top (STT) method [RFC6601]. ALB flooding can be resource intensive, since it requires link bandwidth to carry routing protocol link-state advertisements and processor capacity to process those advertisements; in addition, the overhead of the ALB advertisements and their processing can limit the size of the area and AS. Modeling results suggest that event-dependent TE methods could lead to a reduction in ALB flooding overhead without loss of network throughput performance [TE-QoS-ROUTING].¶

A fully functional TE system is likely to use all aspects of time-dependent, state-dependent, and event-dependent methodologies as described in Section 4.3.1.¶

4.2. Offline versus Online

Traffic engineering requires the computation of routing plans. The computation may be performed offline or online. The computation can be done offline for scenarios where routing plans need not be executed in real time. For example, routing plans computed from forecast information may be computed offline. Typically, offline computation is also used to perform extensive searches on multi-dimensional solution spaces.¶

Online computation is required when the routing plans must adapt to changing network conditions as in state-dependent algorithms. Unlike offline computation (which can be computationally demanding), online computation is geared toward relatively simple and fast calculations to select routes, fine-tune the allocations of resources, and perform load balancing.¶

4.3. Centralized versus Distributed

Under centralized control, there is a central authority that determines routing plans and perhaps other TE control parameters on behalf of each router. The central authority periodically collects network-state information from all routers and sends routing information to the routers. The update cycle for information exchange in both directions is a critical parameter directly impacting the performance of the network being controlled. Centralized control may need high processing power and high bandwidth control channels.¶

Distributed control determines route selection by each router autonomously based on the router's view of the state of the network. The network state information may be obtained by the router using a probing method or distributed by other routers on a periodic basis using link-state advertisements. Network state information may also be disseminated under exception conditions. Examples of protocol extensions used to advertise network link-state information are defined in [RFC5305], [RFC6119], [RFC7471], [RFC8570], and [RFC8571]. See also Section 5.1.3.9.¶

4.4. Local versus Global

Traffic-engineering algorithms may require local and global network-state information.¶

Local information is the state of a portion of the domain. Examples include the bandwidth and packet loss rate of a particular path or the state and capabilities of a network link. Local state information may be sufficient for certain instances of distributed control TE.¶

Global information is the state of the entire TE domain. Examples include a global traffic matrix and loading information on each link throughout the domain of interest. Global state information is typically required with centralized control. Distributed TE systems may also need global information in some cases.¶

4.5. Prescriptive versus Descriptive

TE systems may also be classified as prescriptive or descriptive.¶

Prescriptive traffic engineering evaluates alternatives and recommends a course of action. Prescriptive TE can be further categorized as either corrective or perfective. Corrective TE prescribes a course of action to address an existing or predicted anomaly. Perfective TE prescribes a course of action to evolve and improve network performance even when no anomalies are evident.¶

Descriptive traffic engineering, on the other hand, characterizes the state of the network and assesses the impact of various policies without recommending any particular course of action.¶

4.6. Open-Loop versus Closed-Loop

Open-loop traffic-engineering control is where control action does not use feedback information from the current network state. However, the control action may use its own local information for accounting purposes.¶

Closed-loop traffic-engineering control is where control action utilizes feedback information from the network state. The feedback information may be in the form of current measurement or recent historical records.¶

4.7. Tactical versus Strategic

Tactical traffic engineering aims to address specific performance problems (such as hotspots) that occur in the network from a tactical perspective, without consideration of overall strategic imperatives. Without proper planning and insights, tactical TE tends to be ad hoc in nature.¶

Strategic traffic-engineering approaches the TE problem from a more organized and systematic perspective, taking into consideration the immediate and longer-term consequences of specific policies and actions.¶

5.1. Overview of IETF Projects Related to Traffic Engineering

This subsection reviews a number of IETF activities pertinent to Internet traffic engineering. Some of these technologies are widely deployed, others are mature but have seen less deployment, and some are unproven or are still under development.¶

5.2. Content Distribution

The Internet is dominated by client-server interactions, principally web traffic and multimedia streams, although in the future, more sophisticated media servers may become dominant. The location and performance of major information servers have a significant impact on the traffic patterns within the Internet as well as on the perception of service quality by end users.¶

A number of dynamic load-balancing techniques have been devised to improve the performance of replicated information servers. These techniques can cause spatial traffic characteristics to become more dynamic in the Internet because information servers can be dynamically picked based upon the location of the clients, the location of the servers, the relative utilization of the servers, the relative performance of different networks, and the relative performance of different parts of a network. This process of assignment of distributed servers to clients is called "traffic directing". It is an application-layer function.¶

Traffic-directing schemes that allocate servers in multiple geographically dispersed locations to clients may require empirical network performance statistics to make more effective decisions. In the future, network measurement systems may need to provide this type of information.¶

When congestion exists in the network, traffic-directing and traffic-engineering systems should act in a coordinated manner. This topic is for further study.¶

The issues related to location and replication of information servers, particularly web servers, are important for Internet traffic engineering because these servers contribute a substantial proportion of Internet traffic.¶

6.1. Generic Non-functional Recommendations

The generic non-functional recommendations for Internet traffic engineering are listed in the paragraphs that follow. In a given context, some of these recommendations may be critical while others may be optional. Therefore, prioritization may be required during the development phase of a TE system to tailor it to a specific operational context.¶

Automation:

Whenever feasible, a TE system should automate as many TE functions as possible to minimize the amount of human effort needed to analyze and control operational networks. Automation is particularly important in large-scale public networks because of the high cost of the human aspects of network operations and the high risk of network problems caused by human errors. Automation may additionally benefit from feedback from the network that indicates the state of network resources and the current load in the network. Further, placing intelligence into components of the TE system could enable automation to be more dynamic and responsive to changes in the network.¶

Flexibility:

A TE system should allow for changes in optimization policy. In particular, a TE system should provide sufficient configuration options so that a network administrator can tailor the system to a particular environment. It may also be desirable to have both online and offline TE subsystems that can be independently enabled and disabled. TE systems that are used in multi-class networks should also have options to support class-based performance evaluation and optimization.¶

Interoperability:

Whenever feasible, TE systems and their components should be developed with open standards-based interfaces to allow interoperation with other systems and components.¶

Scalability:

Public networks continue to grow rapidly with respect to network size and traffic volume. Therefore, to remain applicable as the network evolves, a TE system should be scalable. In particular, a TE system should remain functional as the network expands with regard to the number of routers and links and with respect to the number of flows and the traffic volume. A TE system should have a scalable architecture, should not adversely impair other functions and processes in a network element, and should not consume too many network resources when collecting and distributing state information or when exerting control.¶

Security:

Security is a critical consideration in TE systems. Such systems typically exert control over functional aspects of the network to achieve the desired performance objectives. Therefore, adequate measures must be taken to safeguard the integrity of the TE system. Adequate measures must also be taken to protect the network from vulnerabilities that originate from security breaches and other impairments within the TE system.¶

Simplicity:

A TE system should be as simple as possible. Simplicity in user interface does not necessarily imply that the TE system will use naive algorithms. When complex algorithms and internal structures are used, the user interface should hide such complexities from the network administrator as much as possible.¶

Stability:

Stability refers to the resistance of the network to oscillate (flap) in a disruptive manner from one state to another, which may result in traffic being routed first one way and then another without satisfactory resolution of the underlying TE issues and with continued changes that do not settle down. Stability is a very important consideration in TE systems that respond to changes in the state of the network. State-dependent TE methodologies typically include a trade-off between responsiveness and stability. It is strongly recommended that when a trade-off between responsiveness and stability is needed, it should be made in favor of stability (especially in public IP backbone networks).¶

Usability:

Usability is a human aspect of TE systems. It refers to the ease with which a TE system can be deployed and operated. In general, it is desirable to have a TE system that can be readily deployed in an existing network. It is also desirable to have a TE system that is easy to operate and maintain.¶

Visibility:

Mechanisms should exist as part of the TE system to collect statistics from the network and to analyze these statistics to determine how well the network is functioning. Derived statistics (such as traffic matrices, link utilization, latency, packet loss, and other performance measures of interest) that are determined from network measurements can be used as indicators of prevailing network conditions. The capabilities of the various components of the routing system are other examples of status information that should be observable.¶

6.2. Routing Recommendations

Routing control is a significant aspect of Internet traffic engineering. Routing impacts many of the key performance measures associated with networks, such as throughput, delay, and utilization. Generally, it is very difficult to provide good service quality in a wide area network without effective routing control. A desirable TE routing system is one that takes traffic characteristics and network constraints into account during route selection while maintaining stability.¶

Shortest Path First (SPF) IGPs are based on shortest path algorithms and have limited control capabilities for TE [RFC2702] [AWD2]. These limitations include:¶

Because of these limitations, capabilities are needed to enhance the routing function in IP networks. Some of these capabilities are summarized below:¶

6.3. Traffic Mapping Recommendations

Traffic mapping is the assignment of traffic workload onto (pre-established) paths to meet certain requirements. Thus, while constraint-based routing deals with path selection, traffic mapping deals with the assignment of traffic to established paths that may have been generated by constraint-based routing or by some other means. Traffic mapping can be performed by time-dependent or state-dependent mechanisms, as described in Section 4.1.¶

Two important aspects of the traffic mapping function are the ability to establish multiple paths between an originating node and a destination node and the capability to distribute the traffic across those paths according to configured policies. A precondition for this scheme is the existence of flexible mechanisms to partition traffic and then assign the traffic partitions onto the parallel paths (described as "parallel traffic trunks" in [RFC2702]). When traffic is assigned to multiple parallel paths, it is recommended that special care should be taken to ensure proper ordering of packets belonging to the same application (or traffic flow) at the destination node of the parallel paths.¶

Mechanisms that perform the traffic mapping functions should aim to map the traffic onto the network infrastructure to minimize congestion. If the total traffic load cannot be accommodated, or if the routing and mapping functions cannot react fast enough to changing traffic conditions, then a traffic mapping system may use short timescale congestion control mechanisms (such as queue management, scheduling, etc.) to mitigate congestion. Thus, mechanisms that perform the traffic mapping functions complement existing congestion control mechanisms. In an operational network, traffic should be mapped onto the infrastructure such that intra-class and inter-class resource contention are minimized (see Section 2).¶

When traffic mapping techniques that depend on dynamic state feedback (e.g., MPLS Adaptive Traffic Engineering (MATE) [MATE] and suchlike) are used, special care must be taken to guarantee network stability.¶

6.4. Measurement Recommendations

The importance of measurement in TE has been discussed throughout this document. A TE system should include mechanisms to measure and collect statistics from the network to support the TE function. Additional capabilities may be needed to help in the analysis of the statistics. The actions of these mechanisms should not adversely affect the accuracy and integrity of the statistics collected. The mechanisms for statistical data acquisition should also be able to scale as the network evolves.¶

Traffic statistics may be classified according to long-term or short-term timescales. Long-term traffic statistics are very useful for traffic engineering. Long-term traffic statistics may periodically record network workload (such as hourly, daily, and weekly variations in traffic profiles) as well as traffic trends. Aspects of the traffic statistics may also describe class of service characteristics for a network supporting multiple classes of service. Analysis of the long-term traffic statistics may yield other information such as busy-hour characteristics, traffic growth patterns, persistent congestion problems, hotspots, and imbalances in link utilization caused by routing anomalies.¶

A mechanism for constructing traffic matrices for both long-term and short-term traffic statistics should be in place. In multi-service IP networks, the traffic matrices may be constructed for different service classes. Each element of a traffic matrix represents a statistic about the traffic flow between a pair of abstract nodes. An abstract node may represent a router, a collection of routers, or a site in a VPN.¶

Traffic statistics should provide reasonable and reliable indicators of the current state of the network on the short-term scale. Some short-term traffic statistics may reflect link utilization and link congestion status. Examples of congestion indicators include excessive packet delay, packet loss, and high resource utilization. Examples of mechanisms for distributing this kind of information include SNMP, probing tools, FTP, IGP link-state advertisements, NETCONF/RESTCONF, etc.¶

6.5. Policing, Planning, and Access Control

The recommendations in Sections 6.2 and 6.3 may be sub-optimal or even ineffective if the amount of traffic flowing on a route or path exceeds the capacity of the resource on that route or path. Several approaches can be used to increase the performance of TE systems:¶

Combining some elements of all three of these measures is advisable to achieve a better TE system.¶

6.6. Network Survivability

Network survivability refers to the capability of a network to maintain service continuity in the presence of faults. This can be accomplished by promptly recovering from network impairments and maintaining the required QoS for existing services after recovery. Survivability is an issue of great concern within the Internet community due to the demand to carry mission-critical traffic, real-time traffic, and other high-priority traffic over the Internet. Survivability can be addressed at the device level by developing network elements that are more reliable and at the network level by incorporating redundancy into the architecture, design, and operation of networks. It is recommended that a philosophy of robustness and survivability should be adopted in the architecture, design, and operation of TE used to control IP networks (especially public IP networks). Because different contexts may demand different levels of survivability, the mechanisms developed to support network survivability should be flexible so that they can be tailored to different needs. A number of tools and techniques have been developed to enable network survivability, including MPLS Fast Reroute [RFC4090], Topology Independent Loop-free Alternate Fast Reroute for Segment Routing [SR-TI-LFA], RSVP-TE Extensions in Support of End-to-End GMPLS Recovery [RFC4872], and GMPLS Segment Recovery [RFC4873].¶

The impact of service outages varies significantly for different service classes depending on the duration of the outage, which can vary from milliseconds (with minor service impact) to seconds (with possible call drops for IP telephony and session timeouts for connection-oriented transactions) to minutes and hours (with potentially considerable social and business impact). Outages of different durations have different impacts depending on the nature of the traffic flows that are interrupted.¶

Failure protection and restoration capabilities are available in multiple layers as network technologies have continued to evolve. Optical networks are capable of providing dynamic ring and mesh restoration functionality at the wavelength level. At the SONET/SDH layer, survivability capability is provided with Automatic Protection Switching (APS) as well as self-healing ring and mesh architectures. Similar functionality is provided by Layer 2 technologies such as Ethernet.¶

Rerouting is used at the IP layer to restore service following link and node outages. Rerouting at the IP layer occurs after a period of routing convergence, which may require seconds to minutes to complete. Path-oriented technologies such as MPLS [RFC3469] can be used to enhance the survivability of IP networks in a potentially cost-effective manner.¶

An important aspect of multi-layer survivability is that technologies at different layers may provide protection and restoration capabilities at different granularities in terms of timescales and at different bandwidth granularities (from the level of packets to that of wavelengths). Protection and restoration capabilities can also be sensitive to different service classes and different network utility models. Coordinating different protection and restoration capabilities across multiple layers in a cohesive manner to ensure network survivability is maintained at reasonable cost is a challenging task. Protection and restoration coordination across layers may not always be feasible, because networks at different layers may belong to different administrative domains.¶

Some of the general recommendations for protection and restoration coordination are as follows:¶

6.7. Multi-Layer Traffic Engineering

Networks are often implemented as layers. A layer relationship may represent the interaction between technologies (for example, an IP network operated over an optical network) or the relationship between different network operators (for example, a customer network operated over a service provider's network). Note that a multi-layer network does not imply the use of multiple technologies, although some form of encapsulation is often applied.¶

Multi-layer traffic engineering presents a number of challenges associated with scalability and confidentiality. These issues are addressed in [RFC7926], which discusses the sharing of information between domains through policy filters, aggregation, abstraction, and virtualization. That document also discusses how existing protocols can support this scenario with special reference to BGP-LS (see Section 5.1.3.10).¶

PCE (see Section 5.1.3.11) is also a useful tool for multi-layer networks as described in [RFC6805], [RFC8685], and [RFC5623]. Signaling techniques for multi-layer TE are described in [RFC6107].¶

See also Section 6.6 for examination of multi-layer network survivability.¶

6.8. Traffic Engineering in Diffserv Environments

Increasing requirements to support multiple classes of traffic in the Internet, such as best-effort and mission-critical data, call for IP networks to differentiate traffic according to some criteria and to give preferential treatment to certain types of traffic. Large numbers of flows can be aggregated into a few behavior aggregates based on some criteria based on common performance requirements in terms of packet loss ratio, delay, and jitter or in terms of common fields within the IP packet headers.¶

Differentiated Services (Diffserv) [RFC2475] can be used to ensure that SLAs defined to differentiate between traffic flows are met. Classes of service can be supported in a Diffserv environment by concatenating Per-Hop Behaviors (PHBs) along the routing path. A PHB is the forwarding behavior that a packet receives at a Diffserv-compliant node, and it can be configured at each router. PHBs are delivered using buffer-management and packet-scheduling mechanisms and require that the ingress nodes use traffic classification, marking, policing, and shaping.¶

TE can complement Diffserv to improve utilization of network resources. TE can be operated on an aggregated basis across all service classes [RFC3270] or on a per-service-class basis. The former is used to provide better distribution of the traffic load over the network resources (see [RFC3270] for detailed mechanisms to support aggregate TE). The latter case is discussed below since it is specific to the Diffserv environment, with so-called Diffserv-aware traffic engineering [RFC4124].¶

For some Diffserv networks, it may be desirable to control the performance of some service classes by enforcing relationships between the traffic workload contributed by each service class and the amount of network resources allocated or provisioned for that service class. Such relationships between demand and resource allocation can be enforced using a combination of, for example:¶

It may also be desirable to limit the performance impact of high-priority traffic on relatively low-priority traffic. This can be achieved, for example, by controlling the percentage of high-priority traffic that is routed through a given link. Another way to accomplish this is to increase link capacities appropriately so that lower-priority traffic can still enjoy adequate service quality. When the ratio of traffic workload contributed by different service classes varies significantly from router to router, it may not be enough to rely on conventional IGP routing protocols or on TE mechanisms that are not sensitive to different service classes. Instead, it may be desirable to perform TE, especially routing control and mapping functions, on a per-service-class basis. One way to accomplish this in a domain that supports both MPLS and Diffserv is to define class-specific LSPs and to map traffic from each class onto one or more LSPs that correspond to that service class. An LSP corresponding to a given service class can then be routed and protected/restored in a class-dependent manner, according to specific policies.¶

Performing TE on a per-class basis may require per-class parameters to be distributed. It is common to have some classes share some aggregate constraints (e.g., maximum bandwidth requirement) without enforcing the constraint on each individual class. These classes can be grouped into class types, and per-class-type parameters can be distributed to improve scalability. This also allows better bandwidth sharing between classes in the same class type. A class type is a set of classes that satisfy the following two conditions:¶

See [RFC4124] for detailed requirements on Diffserv-aware TE.¶

6.9. Network Controllability

Offline and online (see Section 4.2) TE considerations are of limited utility if the network cannot be controlled effectively to implement the results of TE decisions and to achieve the desired network performance objectives.¶

Capacity augmentation is a coarse-grained solution to TE issues. However, it is simple, may be applied through creating parallel links that form part of an ECMP scheme, and may be advantageous if bandwidth is abundant and cheap. However, bandwidth is not always abundant and cheap, and additional capacity might not always be the best solution. Adjustments of administrative weights and other parameters associated with routing protocols provide finer-grained control, but this approach is difficult to use and imprecise because of the way the routing protocols interact across the network.¶

Control mechanisms can be manual (e.g., static configuration), partially automated (e.g., scripts), or fully automated (e.g., policy-based management systems). Automated mechanisms are particularly useful in large-scale networks. Multi-vendor interoperability can be facilitated by standardized management tools (e.g., YANG models) to support the control functions required to address TE objectives.¶

Network control functions should be secure, reliable, and stable as these are often needed to operate correctly in times of network impairments (e.g., during network congestion or attacks).¶

A.1. RFC 3272

A.2. This Document