Internet Engineering Task Force (IETF) J. Halpern, Ed.
Request for Comments: 7665 Ericsson
Category: Informational C. Pignataro, Ed.
ISSN: 2070-1721 Cisco
Service Function Chaining (SFC) Architecture
This document describes an architecture for the specification,
creation, and ongoing maintenance of Service Function Chains (SFCs)
in a network. It includes architectural concepts, principles, and
components used in the construction of composite services through
deployment of SFCs, with a focus on those to be standardized in the
IETF. This document does not propose solutions, protocols, or
extensions to existing protocols.
Status of This Memo
This document is not an Internet Standards Track specification; it is
published for informational purposes.
This document is a product of the Internet Engineering Task Force
(IETF). It represents the consensus of the IETF community. It has
received public review and has been approved for publication by the
Internet Engineering Steering Group (IESG). Not all documents
approved by the IESG are a candidate for any level of Internet
Standard; see Section 2 of RFC 5741.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
Copyright (c) 2015 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2. Assumptions . . . . . . . . . . . . . . . . . . . . . . . 5
1.3. Specification of Requirements . . . . . . . . . . . . . . 5
1.4. Definition of Terms . . . . . . . . . . . . . . . . . . . 6
2. Architectural Concepts . . . . . . . . . . . . . . . . . . . 8
2.1. Service Function Chains . . . . . . . . . . . . . . . . . 8
2.2. Service Function Chain Symmetry . . . . . . . . . . . . . 9
2.3. Service Function Paths . . . . . . . . . . . . . . . . . 10
2.3.1. Service Function Chains, Service Function Paths, and
Rendered Service Path . . . . . . . . . . . . . . . . 11
3. Architecture Principles . . . . . . . . . . . . . . . . . . . 12
4. Core SFC Architecture Components . . . . . . . . . . . . . . 13
4.1. SFC Encapsulation . . . . . . . . . . . . . . . . . . . . 14
4.2. Service Function (SF) . . . . . . . . . . . . . . . . . . 15
4.3. Service Function Forwarder (SFF) . . . . . . . . . . . . 15
4.3.1. Transport-Derived SFF . . . . . . . . . . . . . . . . 17
4.4. SFC-Enabled Domain . . . . . . . . . . . . . . . . . . . 17
4.5. Network Overlay and Network Components . . . . . . . . . 18
4.6. SFC Proxy . . . . . . . . . . . . . . . . . . . . . . . . 18
4.7. Classification . . . . . . . . . . . . . . . . . . . . . 19
4.8. Reclassification and Branching . . . . . . . . . . . . . 19
4.9. Shared Metadata . . . . . . . . . . . . . . . . . . . . . 20
5. Additional Architectural Concepts . . . . . . . . . . . . . . 21
5.1. The Role of Policy . . . . . . . . . . . . . . . . . . . 21
5.2. SFC Control Plane . . . . . . . . . . . . . . . . . . . . 21
5.3. Resource Control . . . . . . . . . . . . . . . . . . . . 22
5.4. Infinite Loop Detection and Avoidance . . . . . . . . . . 23
5.5. Load-Balancing Considerations . . . . . . . . . . . . . . 23
5.6. MTU and Fragmentation Considerations . . . . . . . . . . 24
5.7. SFC OAM . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.8. Resilience and Redundancy . . . . . . . . . . . . . . . . 26
6. Security Considerations . . . . . . . . . . . . . . . . . . . 27
7. References . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.1. Normative References . . . . . . . . . . . . . . . . . . 29
7.2. Informative References . . . . . . . . . . . . . . . . . 29
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . 30
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 32
The delivery of end-to-end services often requires various service
functions. These include traditional network service functions such
as firewalls and traditional IP Network Address Translators (NATs),
as well as application-specific functions. The definition and
instantiation of an ordered set of service functions and subsequent
"steering" of traffic through them is termed Service Function
This document describes an architecture used for the creation and
ongoing maintenance of Service Function Chains (SFCs) in a network.
It includes architectural concepts, principles, and components, with
a focus on those to be standardized in the IETF. SFCs enable
composite services that are constructed from one or more service
An overview of the issues associated with the deployment of end-to-
end service function chains, abstract sets of service functions and
their ordering constraints that create a composite service, and the
subsequent "steering" of traffic flows through said service
functions, is described in [RFC7498].
The current service function deployment models are relatively static,
coupled to network topology and physical resources, greatly reducing
or eliminating the ability of an operator to introduce new services
or dynamically create service function chains. This architecture
presents a model addressing the problematic aspects of existing
service deployments, including topological independence and
This document defines the architecture for Service Function Chaining
(SFC) as standardized in the IETF. The SFC architecture is
predicated on topological independence from the underlying forwarding
In this architecture, packets are classified on ingress for handling
by the required set of Service Functions (SFs) in the SFC-enabled
domain and are then forwarded through that set of functions for
processing by each function in turn. Packets may be reclassified as
a result of this processing.
The architecture described in this document is independent of the
planned usage of the network and deployment context and thus, for
example, is applicable to both fixed and mobile networks as well as
being useful in many data center applications.
The architecture described herein is assumed to be applicable to a
single network administrative domain. While it is possible for the
architectural principles and components to be applied to inter-domain
SFCs, these are left for future study.
The following assumptions are made:
o There is no standard definition or characterization applicable to
all SFs, and thus the architecture considers each SF as an opaque
o There is no global or standard list of SFs enabled in a given
administrative domain. The set of SFs enabled in a given domain
is a function of the currently active services that may vary with
time and according to the networking environment.
o There is no global or standard SF chaining logic. The ordered set
of SFs that needs to be applied to deliver a given service is
specific to each administrative entity.
o The chaining of SFs and the criteria to invoke them are specific
to each administrative entity that operates an SF-enabled domain.
o Several SF chaining policies can be simultaneously applied within
an administrative domain to meet various business requirements.
o The underlay is assumed to provide the necessary connectivity to
interconnect the Service Function Forwarders (SFFs; see
Section 1.4), but the architecture places no constraints on how
that connectivity is realized other than it have the required
bandwidth, latency, and jitter to support the SFC.
o No assumption is made on how Forwarding Information Bases (FIBs)
and Routing Information Bases (RIBs) of involved nodes are
o How to bind traffic to a given SF chain is policy-based.
1.3. Specification of Requirements
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 [RFC2119].
1.4. Definition of Terms
Network Service: An offering provided by an operator that is
delivered using one or more service functions. This may also be
referred to as a "composite service". The term "service" is
used to denote a "network service" in the context of this
Note: Beyond this document, the term "service" is overloaded
with varying definitions. For example, to some a service is an
offering composed of several elements within the operator's
network, whereas for others a service, or more specifically a
network service, is a discrete element such as a "firewall".
Traditionally, such services (in the latter sense) host a set of
service functions and have a network locator where the service
Classification: Locally instantiated matching of traffic flows
against policy for subsequent application of the required set of
network service functions. The policy may be customer/network/
Classifier: An element that performs Classification.
Service Function Chain (SFC): A service function chain defines an
ordered set of abstract service functions and ordering
constraints that must be applied to packets and/or frames and/or
flows selected as a result of classification. An example of an
abstract service function is "a firewall". The implied order
may not be a linear progression as the architecture allows for
SFCs that copy to more than one branch, and also allows for
cases where there is flexibility in the order in which service
functions need to be applied. The term "service chain" is often
used as shorthand for service function chain.
Service Function (SF): A function that is responsible for specific
treatment of received packets. A Service Function can act at
various layers of a protocol stack (e.g., at the network layer
or other OSI layers). As a logical component, a service
function can be realized as a virtual element or be embedded in
a physical network element. One or more Service Functions can
be embedded in the same network element. Multiple occurrences
of the service function can exist in the same administrative
One or more service functions can be involved in the delivery of
added-value services. A non-exhaustive list of abstract service
functions includes: firewalls, WAN and application acceleration,
Deep Packet Inspection (DPI), Lawful Intercept (LI), server load
balancing, NAT44 [RFC3022], NAT64 [RFC6146], NPTv6 [RFC6296],
HOST_ID injection, HTTP Header Enrichment functions, and TCP
An SF may be SFC encapsulation aware (that is, it receives and
acts on information in the SFC encapsulation) or unaware (in
which case, data forwarded to the SF does not contain the SFC
encapsulation). This is often referred to as "SFC aware" and
"SFC unaware", respectively.
Service Function Forwarder (SFF): A service function forwarder is
responsible for forwarding traffic to one or more connected
service functions according to information carried in the SFC
encapsulation, as well as handling traffic coming back from the
SF. Additionally, an SFF is responsible for delivering traffic
to a classifier when needed and supported, transporting traffic
to another SFF (in the same or different type of overlay), and
terminating the Service Function Path (SFP).
Metadata: Provides the ability to exchange context information
between classifiers and SFs, and among SFs.
Service Function Path (SFP): The service function path is a
constrained specification of where packets assigned to a certain
service function path must go. While it may be so constrained
as to identify the exact locations, it can also be less
specific. The SFP provides a level of indirection between the
fully abstract notion of service chain as a sequence of abstract
service functions to be delivered, and the fully specified
notion of exactly which SFF/SFs the packet will visit when it
actually traverses the network. By allowing the control
components to specify this level of indirection, the operator
may control the degree of SFF/SF selection authority that is
delegated to the network.
SFC Encapsulation: The SFC encapsulation provides, at a minimum, SFP
identification, and is used by the SFC-aware functions, such as
the SFF and SFC-aware SFs. The SFC encapsulation is not used
for network packet forwarding. In addition to SFP
identification, the SFC encapsulation carries metadata including
data-plane context information.
Rendered Service Path (RSP): Within an SFP, packets themselves are
of course transmitted from and to specific places in the
network, visiting a specific sequence of SFFs and SFs. This
sequence of actual visits by a packet to specific SFFs and SFs
in the network is known as the Rendered Service Path (RSP).
This definition is included here for use by later documents,
such as when solutions may need to discuss the actual sequence
of locations the packets visit.
SFC-Enabled Domain: A network or region of a network that implements
SFC. An SFC-enabled domain is limited to a single network
SFC Proxy: Removes and inserts SFC encapsulation on behalf of an
SFC-unaware service function. SFC proxies are logical elements.
2. Architectural Concepts
The following sections describe the foundational concepts of service
function chaining and the SFC architecture.
Service function chaining enables the creation of composite (network)
services that consist of an ordered set of SFs that must be applied
to packets and/or frames and/or flows selected as a result of
classification. Each SF is referenced using an identifier that is
unique within an SF-enabled domain.
Service function chaining is a concept that provides for more than
just the application of an ordered set of SFs to selected traffic;
rather, it describes a method for deploying SFs in a way that enables
dynamic ordering and topological independence of those SFs as well as
the exchange of metadata between participating entities.
2.1. Service Function Chains
In most networks, services are constructed as abstract sequences of
SFs that represent SFCs. At a high level, an SFC is an abstracted
view of a service that specifies the set of required SFs as well as
the order in which they must be executed. Graphs, as illustrated in
Figure 1, define an SFC, where each graph node represents the
required existence of at least one abstract SF. Such graph nodes
(SFs) can be part of zero, one, or many SFCs. A given graph node
(SF) can appear one time or multiple times in a given SFC.
SFCs can start from the origination point of the service function
graph (i.e., node 1 in Figure 1), or from any subsequent node in the
graph. As shown, SFs may therefore become branching nodes in the
graph, with those SFs selecting edges that move traffic to one or
more branches. The top and middle graphs depict such a case, where a
second classification event occurs after node 2, and a new graph is
selected (i.e., node 3 instead of node 6). The bottom graph
highlights the concept of a cycle, in which a given SF (e.g., node 7
in the depiction) can be visited more than once within a given
service chain. An SFC can have more than one terminus.
,-+-. ,---. ,---. ,---.
/ \ / \ / \ / \
( 1 )+--->( 2 )+---->( 6 )+---->( 8 )
\ / \ / \ / \ /
`---' `---' `---' `---'
,-+-. ,---. ,---. ,---. ,---.
/ \ / \ / \ / \ / \
( 1 )+--->( 2 )+---->( 3 )+---->( 7 )+---->( 9 )
\ / \ / \ / \ / \ /
`---' `---' `---' `---' `---'
,-+-. ,---. ,---. ,---. ,---.
/ \ / \ / \ / \ / \
( 1 )+--->( 7 )+---->( 8 )+---->( 4 )+---->( 7 )
\ / \ / \ / \ / \ /
`---' `---' `---' `---' `---'
Figure 1: Service Function Chain Graphs
The concepts of classification, reclassification, and branching are
covered in subsequent sections of this architecture (see Sections 4.7
2.2. Service Function Chain Symmetry
SFCs may be unidirectional or bidirectional. A unidirectional SFC
requires that traffic be forwarded through the ordered SFs in one
direction (sf1 -> sf2 -> sf3), whereas a bidirectional SFC requires a
symmetric path (sf1 -> sf2 -> sf3 and sf3 -> sf2 -> sf1), and in
which the SF instances are the same in opposite directions. A hybrid
SFC has attributes of both unidirectional and bidirectional SFCs;
that is to say some SFs require symmetric traffic, whereas other SFs
do not process reverse traffic or are independent of the
corresponding forward traffic.
SFCs may contain cycles; that is traffic may need to traverse one or
more SFs within an SFC more than once. Solutions will need to ensure
suitable disambiguation for such situations.
The architectural allowance that is made for SFPs that delegate
choice to the network for which SFs and/or SFFs a packet will visit
creates potential issues here. A solution that allows such
delegation needs to also describe how the solution ensures that those
service chains requiring service function chain symmetry can achieve
Further, there are state trade-offs in symmetry. Symmetry may be
realized in several ways depending on the SFF and classifier
functionality. In some cases, "mirrored" classification (i.e., from
Source to Destination and from Destination to Source) policy may be
deployed, whereas in others shared state between classifiers may be
used to ensure that symmetric flows are correctly identified, then
steered along the required SFP. At a high level, there are various
common cases. In a non-exhaustive way, there can be for example:
o A single classifier (or a small number of classifiers), in which
case both incoming and outgoing flows could be recognized at the
same classifier, so the synchronization would be feasible by
internal mechanisms internal to the classifier.
o Stateful classifiers where several classifiers may be clustered
and share state.
o Fully distributed classifiers, where synchronization needs to be
provided through unspecified means.
o A classifier that learns state from the egress packets/flows that
is then used to provide state for the return packets/flow.
o Symmetry may also be provided by stateful forwarding logic in the
SFF in some implementations.
This is a non-comprehensive list of common cases.
2.3. Service Function Paths
A Service Function Path (SFP) is a mechanism used by service chaining
to express the result of applying more granular policy and
operational constraints to the abstract requirements of a service
chain (SFC). This architecture does not mandate the degree of
specificity of the SFP. Architecturally, within the same SFC-enabled
domain, some SFPs may be fully specified, selecting exactly which SFF
and which SF are to be visited by packets using that SFP, while other
SFPs may be quite vague, deferring to the SFF the decisions about the
exact sequence of steps to be used to realize the SFC. The
specificity may be anywhere in between these extremes.
As an example of such an intermediate specificity, there may be two
SFPs associated with a given SFC, where one SFP specifies that any
order of SFF and SF may be used as long as it is within Data Center
1, and where the second SFP allows the same latitude, but only within
Data Center 2.
Thus, the policies and logic of SFP selection or creation (depending
upon the solution) produce what may be thought of as a constrained
version of the original SFC. Since multiple policies may apply to
different traffic that uses the same SFC, it also follows that there
may be multiple SFPs associated with a single SFC.
The architecture allows for the same SF to be reachable through
multiple SFFs. In these cases, some SFPs may constrain which SFF is
used to reach which SF, while some SFPs may leave that decision to
the SFF itself.
Further, the architecture allows for two or more SFs to be attached
to the same SFF, and possibly connected via internal means allowing
more effective communication. In these cases, some solutions or
deployments may choose to use some form of internal inter-process or
inter-VM messaging (communication behind the virtual switching
element) that is optimized for such an environment. This must be
coordinated with the SFF so that it can properly perform its job.
Implementation details of such mechanisms are considered out of scope
for this document, and can include a spectrum of methods: for
example, situations including all next-hops explicitly, others where
a list of possible next-hops is provided and the selection is local,
or cases with just an identifier, where all resolution is local.
This architecture also allows the same SF to be part of multiple
2.3.1. Service Function Chains, Service Function Paths, and Rendered
As an example of this progressive refinement, consider a Service
Function Chain (SFC) that states that packets using this chain should
be delivered to a firewall and a caching engine.
A Service Function Path (SFP) could refine this, considering that
this architecture does not mandate the degree of specificity an SFP
has to have. It might specify that the firewall and caching engine
are both to be in a specific data center (e.g., in DC1), or it might
specify exactly which instance of each firewall and caching engine is
to be used.
The Rendered Service Path (RSP) is the actual sequence of SFFs and
SFs that the packets will actually visit. So if the SFP picked the
DC, the RSP would be more specific.
3. Architecture Principles
Service function chaining is predicated on several key architectural
1. Topological independence: No changes to the underlay network
forwarding topology -- implicit, or explicit -- are needed to
deploy and invoke SFs or SFCs.
2. Plane separation: Dynamic realization of SFPs is separated from
packet handling operations (e.g., packet forwarding).
3. Classification: Traffic that satisfies classification rules is
forwarded according to a specific SFP. For example,
classification can be as simple as an explicit forwarding entry
that forwards all traffic from one address into the SFP.
Multiple classification points are possible within an SFC (i.e.,
forming a service graph), thus enabling changes/updates to the
SFC by SFs.
Classification can occur at varying degrees of granularity; for
example, classification can use a 5-tuple, a transport port or
set of ports, part of the packet payload, it can be the result of
high-level inspections, or it can come from external systems.
4. Shared Metadata: Metadata/context data can be shared amongst SFs
and classifiers, between SFs, and between external systems and
SFs (e.g., orchestration).
One use of metadata is to provide and share the result of
classification (that occurs within the SFC-enabled domain, or
external to it) along an SFP. For example, an external
repository might provide user/subscriber information to a service
chain classifier. This classifier could in turn impose that
information in the SFC encapsulation for delivery to the
requisite SFs. The SFs could in turn utilize the user/subscriber
information for local policy decisions. Metadata can also share
SF output along the SFP.
5. Service definition independence: The SFC architecture does not
depend on the details of SFs themselves.
6. Service function chain independence: The creation, modification,
or deletion of an SFC has no impact on other SFCs. The same is
true for SFPs.
7. Heterogeneous control/policy points: The architecture allows SFs
to use independent mechanisms (out of scope for this document) to
populate and resolve local policy and (if needed) local
4. Core SFC Architecture Components
The SFC Architecture is built out of architectural building blocks
that are logical components; these logical components are
classifiers, Service Function Forwarders (SFFs), the Service
Functions (SFs) themselves, and SFC proxies. While this architecture
describes functionally distinct logical components and promotes
transport independence, they could be realized and combined in
various ways in deployed products, and could be combined with an
They are interconnected using the SFC encapsulation. This results in
a high-level logical architecture of an SFC-enabled domain that
o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. +--------------+ +------------------~~~
. | Service | SFC | Service +---+ +---+
. |Classification| Encapsulation | Function |sf1|...|sfn|
+---->| Function |+---------------->| Path +---+ +---+
. +--------------+ +------------------~~~
. SFC-enabled Domain
o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Figure 2: Service Function Chain Architecture
The following subsections provide details on each logical component
that form the basis of the SFC architecture. A detailed overview of
how some of these architectural components interact is provided in
| SFC-aware | | SFC-unaware |
|Service Function| |Service Function|
SFC Encapsulation No SFC Encapsulation
| SFC |
+---------+ +----------------+ Encapsulation +---------+
|SFC-Aware|-----------------+ \ +------------|SFC Proxy|
| SF | ... ----------+ \ \ / +---------+
+---------+ \ \ \ /
| SF Forwarder |
| (SFF) |
... SFC-enabled Domain ...
Network Overlay Transport
| Network |
Figure 3: SFC Architecture Components After Initial Classification
Please note that the depiction in Figure 3 shows packets after
initial classification, and therefore includes the SFC encapsulation.
Although not included in Figure 3, the classifier is an SFC
4.1. SFC Encapsulation
The SFC encapsulation enables service function path selection. It
also enables the sharing of metadata/context information when such
metadata exchange is required.
The SFC encapsulation carries explicit information used to identify
the SFP. However, the SFC encapsulation is not a transport
encapsulation itself: it is not used to forward packets within the
network fabric. If packets need to flow between separate physical
platforms, the SFC encapsulation relies on an outer network
transport. Transit forwarders -- such as router and switches --
forward SFC encapsulated packets based on the outer (non-SFC)
One of the key architecture principles of SFC is that the SFC
encapsulation remain transport independent. As such, any network
transport protocol may be used to carry the SFC encapsulated traffic.
4.2. Service Function (SF)
The concept of an SF evolves; rather than being viewed as a bump in
the wire, an SF becomes a resource within a specified administrative
domain that is available for consumption as part of a composite
service. SFs send/receive data to/from one or more SFFs. SFC-aware
SFs receive this traffic with the SFC encapsulation.
While the SFC architecture defines the concept and specifies some
characteristics of a new encapsulation -- the SFC encapsulation --
and several logical components for the construction of SFCs, existing
SF implementations may not have the capabilities to act upon or fully
integrate with the new SFC encapsulation. In order to provide a
mechanism for such SFs to participate in the architecture, an SFC
proxy function is defined (see Section 4.6). The SFC proxy acts as a
gateway between the SFC encapsulation and SFC-unaware SFs. The
integration of SFC-unaware service functions is discussed in more
detail in the SFC proxy section.
This architecture allows an SF to be part of multiple SFPs and SFCs.
4.3. Service Function Forwarder (SFF)
The SFF is responsible for forwarding packets and/or frames received
from the network to one or more SFs associated with a given SFF using
information conveyed in the SFC encapsulation. Traffic from SFs
eventually returns to the same SFF, which is responsible for
injecting traffic back onto the network. Some SFs, such as
firewalls, could also consume a packet.
The collection of SFFs and associated SFs creates a service-plane
overlay in which SFC-aware SFs, as well as SFC-unaware SFs reside.
Within this service plane, the SFF component connects different SFs
that form a service function path.
SFFs maintain the requisite SFP forwarding information. SFP
forwarding information is associated with a service path identifier
that is used to uniquely identify an SFP. The service forwarding
state enables an SFF to identify which SFs of a given SFP should be
applied, and in what order, as traffic flows through the associated
SFP. While there may appear to the SFF to be only one available way
to deliver the given SF, there may also be multiple choices allowed
by the constraints of the SFP.
If there are multiple choices, the SFF needs to preserve the property
that all packets of a given flow are handled the same way, since the
SF may well be stateful. Additionally, the SFF may preserve the
handling of packets based on other properties on top of a flow, such
as a subscriber, session, or application instance identification.
The SFF also has the information that allows it to forward packets to
the next SFF after applying local service functions. Again, while
there may be only a single choice available, the architecture allows
for multiple choices for the next SFF. As with SFs, the solution
needs to operate such that the behavior with regard to specific flows
(see the Rendered Service Path) is stable. The selection of
available SFs and next SFFs may be interwoven when an SFF supports
multiple distinct service functions and the same service function is
available at multiple SFFs. Solutions need to be clear about what is
allowed in these cases.
Even when the SFF supports and utilizes multiple choices, the
decision as to whether to use flow-specific mechanisms or coarser-
grained means to ensure that the behavior of specific flows is stable
is a matter for specific solutions and specific implementations.
The SFF component has the following primary responsibilities:
1. SFP forwarding: Traffic arrives at an SFF from the network. The
SFF determines the appropriate SF the traffic should be forwarded
to via information contained in the SFC encapsulation. After SF
processing, the traffic is returned to the SFF, and, if needed,
is forwarded to another SF associated with that SFF. If there is
another non-local (i.e., different SFF) hop in the SFP, the SFF
further encapsulates the traffic in the appropriate network
transport protocol and delivers it to the network for delivery to
the next SFF along the path. Related to this forwarding
responsibility, an SFF should be able to interact with metadata.
2. Terminating SFPs: An SFC is completely executed when traffic has
traversed all required SFs in a chain. When traffic arrives at
the SFF after the last SF has finished processing it, the final
SFF knows from the service forwarding state that the SFC is
complete. The SFF removes the SFC encapsulation and delivers the
packet back to the network for forwarding.
3. Maintaining flow state: In some cases, the SFF may be stateful.
It creates flows and stores flow-centric information. This state
information may be used for a range of SFP-related tasks such as
ensuring consistent treatment of all packets in a given flow,
ensuring symmetry, or for state-aware SFC Proxy functionality
(see Section 4.8).
4.3.1. Transport-Derived SFF
SFP forwarding, as described above, directly depends upon the use of
the service path information contained in the SFC encapsulation.
However, existing implementations may not be able to act on the SFC
encapsulation. These platforms may opt to use existing transport
information, if it can be arranged, to provide explicit service path
This results in the same architectural behavior and meaning for SFP
forwarding and service function paths. It is the responsibility of
the control components to ensure that the transport path executed in
such a case is fully aligned with the path identified by the
information in the service chaining encapsulation.
4.4. SFC-Enabled Domain
Specific features may need to be enforced at the boundaries of an
SFC-enabled domain, for example to avoid leaking SFC information.
Using the term "node" to refer generically to an entity that is
performing a set of functions, in this context, an SFC boundary node
denotes a node that connects one SFC-enabled domain to a node either
located in another SFC-enabled domain or in a domain that is SFC-
An SFC boundary node can act as egress or ingress. An SFC Egress
Node denotes an SFC boundary node that handles traffic leaving the
SFC-enabled domain the Egress Node belongs to. Such a node is
required to remove any information specific to the SFC Domain,
typically the SFC encapsulation. Further, from a privacy
perspective, an SFC Egress Node is required to ensure that any
sensitive information added as part of SFC gets removed. In this
context, information may be sensitive due to network concerns or end-
customer concerns. An SFC Ingress Node denotes an SFC boundary node
that handles traffic entering the SFC-enabled domain. In most
solutions and deployments this will need to include a classifier, and
will be responsible for adding the SFC encapsulation to the packet.
An SFC Proxy and corresponding SFC-unaware service function (see
Figure 3) are inside the SFC-enabled domain.
4.5. Network Overlay and Network Components
Underneath the SFF there are components responsible for performing
the transport (overlay) forwarding. They do not consult the SFC
encapsulation or inner payload for performing this forwarding. They
only consult the outer-transport encapsulation for the transport
4.6. SFC Proxy
In order for the SFC architecture to support SFC-unaware SFs (e.g.,
legacy service functions) a logical SFC proxy function may be used.
This function sits between an SFF and one or more SFs to which the
SFF is directing traffic (see Figure 3).
The proxy accepts packets from the SFF on behalf of the SF. It
removes the SFC encapsulation, and then uses a local attachment
circuit to deliver packets to SFC-unaware SFs. It also receives
packets back from the SF, reapplies the SFC encapsulation, and
returns them to the SFF for processing along the service function
Thus, from the point of view of the SFF, the SFC proxy appears to be
part of an SFC-aware SF.
Communication details between the SFF and the SFC Proxy are the same
as those between the SFF and an SFC-aware SF. The details of that
are not part of this architecture. The details of the communication
methods over the local attachment circuit between the SFC proxy and
the SFC-unaware SF are dependent upon the specific behaviors and
capabilities of that SFC-unaware SF, and thus are also out of scope
for this architecture.
Specifically, for traffic received from the SFF intended for the SF
the proxy is representing, the SFC proxy:
o Removes the SFC encapsulation from SFC encapsulated packets.
o Identifies the required SF to be applied based on available
information including that carried in the SFC encapsulation.
o Selects the appropriate outbound local attachment circuit through
which the next SF for this SFP is reachable. This is derived from
the identification of the SF carried in the SFC encapsulation, and
may include local techniques. Examples of a local attachment
circuit include, but are not limited to, VLAN, IP-in-IP, Layer 2
Tunneling Protocol version 3 (L2TPv3), Generic Routing
Encapsulation (GRE), and Virtual eXtensible Local Area Network
o Forwards the original payload via the selected local attachment
circuit to the appropriate SF.
When traffic is returned from the SF:
o Applies the required SFC encapsulation. The determination of the
encapsulation details may be inferred by the local attachment
circuit through which the packet and/or frame was received, or via
packet classification, or other local policy. In some cases,
packet ordering or modification by the SF may necessitate
additional classification in order to reapply the correct SFC
o Delivers the packet with the SFC encapsulation to the SFF, as
would happen with packets returned from an SFC-aware SF.
Traffic from the network that satisfies classification criteria is
directed into an SFP and forwarded to the requisite service
function(s). Classification is handled by a service classification
function; initial classification occurs at the ingress to the SFC
domain. The granularity of the initial classification is determined
by the capabilities of the classifier and the requirements of the SFC
policy. For instance, classification might be relatively coarse: all
packets from this port are subject to SFC policy X and directed into
SFP A, or quite granular: all packets matching this 5-tuple are
subject to SFC policy Y and directed into SFP B.
As a consequence of the classification decision, the appropriate SFC
encapsulation is imposed on the data, and a suitable SFP is selected
or created. Classification results in attaching the traffic to a
4.8. Reclassification and Branching
The SFC architecture supports reclassification (or non-initial
classification) as well. As packets traverse an SFP,
reclassification may occur -- typically performed by a classification
function co-resident with a service function. Reclassification may
result in the selection of a new SFP, an update of the associated
metadata, or both. This is referred to as "branching".
For example, an initial classification results in the selection of
SFP A: DPI_1 --> SLB_8. However, when the DPI service function is
executed, attack traffic is detected at the application layer. DPI_1
reclassifies the traffic as attack and alters the service path to SFP
B, to include a firewall for policy enforcement: dropping the
traffic: DPI_1 --> FW_4. Subsequent to FW_4, surviving traffic would
be returned to the original SFF. In this simple example, the DPI
service function reclassifies the traffic based on local application
layer classification capabilities (that were not available during the
initial classification step).
When traffic arrives after being steered through an SFC-unaware SF,
the SFC Proxy must perform reclassification of traffic to determine
the SFP. The SFC Proxy is concerned with re-attaching information
for SFC-unaware SFs, and a stateful SFC Proxy simplifies such
classification to a flow lookup.
4.9. Shared Metadata
Sharing metadata allows the network to provide network-derived
information to the SFs, SF-to-SF information exchange, and the
sharing of service-derived information to the network. Some SFCs may
not require metadata exchange. SFC infrastructure enables the
exchange of this shared data along the SFP. The shared metadata
serves several possible roles within the SFC architecture:
o Allows elements that typically operate independently (e.g., as
"ships in the night") to exchange information.
o Encodes information about the network and/or data for subsequent
use within the SFP.
o Creates an identifier used for policy binding by SFs.
Context information can be derived in several ways:
o External sources
o Network node classification
o Service function classification
5. Additional Architectural Concepts
There are a number of issues that solutions need to address, and that
the architecture informs but does not determine. This section lays
out some of those concepts.
5.1. The Role of Policy
Much of the behavior of service chains is driven by operator and per-
customer policy. This architecture is structured to isolate the
policy interactions from the data plane and control logic.
Specifically, it is assumed that the service chaining control plane
creates the service paths. The service chaining data plane is used
to deliver the classified packets along the service chains to the
intended service functions.
Policy, in contrast, interacts with the system in other places.
Policies and policy engines may monitor service functions to decide
if additional (or fewer) instances of services are needed. When
applicable, those decisions may in turn result in interactions that
direct the control logic to change the SFP placement or packet
Similarly, operator service policy, often managed by Operations or
Business Support Systems (OSS or BSS), will frequently determine what
service functions are available. Operator service policies also
determine which sequences of functions are valid and are to be used
or made available.
The offering of service chains to customers, and the selection of
which service chain a customer wishes to use, are driven by a
combination of operator and customer policies using appropriate
portals in conjunction with the OSS and BSS tools. These selections
then drive the service chaining control logic, which in turn
establishes the appropriate packet classification rules.
5.2. SFC Control Plane
The SFC control plane is part of the overall SFC architecture, and
this section describes its high-level functions. However, the
detailed definition of the SFC control plane is outside the scope of
The SFC control plane is responsible for constructing SFPs,
translating SFCs to forwarding paths, and propagating path
information to participating nodes to achieve requisite forwarding
behavior to construct the service overlay. For instance, an SFC
construction may be static; selecting exactly which SFFs and which
SFs from those SFFs are to be used, or it may be dynamic, allowing
the network to perform some or all of the choices of SFF or SF to use
to deliver the selected service chain within the constraints
represented by the service path.
In the SFC architecture, SFs are resources; the control plane manages
and communicates their capabilities, availability, and location in
fashions suitable for the transport and SFC operations in use. The
control plane is also responsible for the creation of the context
(see below). The control plane may be distributed (using new or
existing control-plane protocols), or be centralized, or a
combination of the two.
The SFC control plane provides the following functionality:
1. An SFC-enabled domain wide view of all available service function
resources as well as the network locators through which they are
2. Uses SFC policy to construct service function chains, and
3. Selection of specific SFs for a requested SFC, either statically
(using specific SFs) or dynamically (using service explicit SFs
at the time of delivering traffic to them).
4. Provides requisite SFC data-plane information to the SFC
architecture components, most notably the SFF.
5. Provides the metadata and usage information classifiers need so
that they in turn can provide this metadata for appropriate
packets in the data plane.
6. When needed, provide information including policy information to
other SFC elements to be able to properly interpret metadata.
5.3. Resource Control
The SFC system may be responsible for managing all resources
necessary for the SFC components to function. This includes network
constraints used to plan and choose network path(s) between service
function forwarders, network communication paths between service
function forwarders and their attached service functions,
characteristics of the nodes themselves such as memory, number of
virtual interfaces, routes, and instantiation, configuration, and
deletion of SFs.
The SFC system will also be required to reflect policy decisions
about resource control, as expressed by other components in the
While all of these aspects are part of the overall system, they are
beyond the scope of this architecture.
5.4. Infinite Loop Detection and Avoidance
This SFC architecture is predicated on topological independence from
the underlying forwarding topology. Consequently, a service topology
is created by service function paths or by the local decisions of the
service function forwarders based on the constraints expressed in the
SFP. Due to the overlay constraints, the packet-forwarding path may
need to visit the same SFF multiple times, and in some less common
cases may even need to visit the same SF more than once. The Service
Chaining solution needs to permit these limited and policy-compliant
loops. At the same time, the solutions must ensure that indefinite
and unbounded loops cannot be formed, as such would consume unbounded
resources without delivering any value.
In other words, this architecture requires the solution to prevent
infinite service function loops, even when service functions may be
invoked multiple times in the same SFP.
5.5. Load-Balancing Considerations
Supporting function elasticity and high-availability should not
overly complicate SFC or lead to unnecessary scalability problems.
In the simplest case, where there is only a single function in the
SFP (the next hop is either the destination address of the flow or
the appropriate next hop to that destination), one could argue that
there may be no need for SFC.
In the cases where the classifier is separate from the single
function or a function at the terminal address may need a sub-prefix
(e.g., finer-grained address information) or per-subscriber metadata,
a single SFP exists (i.e., the metadata changes but the SFP does
not), regardless of the number of potential terminal addresses for
the flow. This is the case of the simple load balancer. See
+---+ +---++--->web server
+---+ +---++--->web server
Figure 4: Simple Load Balancing
By extrapolation, in the case where intermediary functions within a
chain had similar "elastic" behaviors, we do not need separate chains
to account for this behavior -- as long as the traffic coalesces to a
common next-hop after the point of elasticity.
In Figure 5, we have a chain of five service functions between the
traffic source and its destination.
+---+ +---+ +---+ +---+ +---+ +---+
|sf2| |sf2| |sf3| |sf3| |sf4| |sf4|
+---+ +---+ +---+ +---+ +---+ +---+
| | | | | |
+---+ +---+ +---+ +---+ +---+
+---+ +---+ +---+ +---+ +---+
+ + +
| | |
+---+ +---+ +---+
|sf1| |sf3| |sf5|
+---+ +---+ +---+
Figure 5: Load Balancing
This would be represented as one service function path:
sf1 -> sf2 -> sf3 -> sf4 -> sf5. The SFF is a logical element, which
may be made up of one or multiple components. In this architecture,
the SFF may handle load distribution based on policy.
It can also be seen in the above that the same service function may
be reachable through multiple SFFs, as discussed earlier. The
selection of which SFF to use to reach sf3 may be made by the control
logic in defining the SFP, or may be left to the SFFs themselves,
depending upon policy, solution, and deployment constraints. In the
latter case, it needs to be assured that exactly one SFF takes
responsibility to steer traffic through sf3.
5.6. MTU and Fragmentation Considerations
This architecture prescribes that additional information be added to
packets to identify service function paths and often to represent
metadata. It also envisions adding transport information to carry
packets along service function paths, at least between service
function forwarders. This added information increases the size of
the packet to be carried by service chaining. Such additions could
potentially increase the packet size beyond the MTU supported on some
or all of the media used in the service chaining domain.
Such packet size increases can thus cause operational MTU problems.
Requiring fragmentation and reassembly in an SFF would be a major
processing increase and might be impossible with some transports.
Expecting service functions to deal with packets fragmented by the
SFC function might be onerous even when such fragmentation was
possible. Thus, at the very least, solutions need to pay attention
to the size cost of their approach. There may be alternative or
additional means available, although any solution needs to consider
These considerations apply to any generic architecture that increases
the header size. There are also more specific MTU considerations:
Effects on Path MTU Discovery (PMTUD) as well as deployment
considerations. Deployments within a single administrative control
or even a single data center complex can afford more flexibility in
dealing with larger packets, and deploying existing mitigations that
decrease the likelihood of fragmentation or discard.
5.7. SFC OAM
Operations, Administration, and Maintenance (OAM) tools are an
integral part of the architecture. These serve various purposes,
including fault detection and isolation, and performance management.
For example, there are many advantages of SFP liveness detection,
including status reporting, support for resiliency operations and
policies, and an enhanced ability to balance load.
Service function paths create a services topology, and OAM performs
various functions within this service layer. Furthermore, SFC OAM
follows the same architectural principles of SFC in general. For
example, topological independence (including the ability to run OAM
over various overlay technologies) and classification-based policy.
We can subdivide the SFC OAM architecture in two parts:
o In-band: OAM packets follow the same path and share fate with user
packets, within the service topology. For this, they also follow
the architectural principle of consistent policy identifiers, and
use the same path IDs as the service chain data packets. Load
balancing and SFC encapsulation with packet forwarding are
particularly important here.
o Out-of-band: Reporting beyond the actual data plane. An
additional layer beyond the data-plane OAM allows for additional
alerting and measurements.
This architecture prescribes end-to-end SFP OAM functions, which
implies SFF understanding of whether an in-band packet is an OAM or
user packet. However, service function validation is outside of the
scope of this architecture, and application-level OAM is not what
this architecture prescribes.
Some of the detailed functions performed by SFC OAM include fault
detection and isolation in a service function path or a service
function, verification that connectivity using SFPs is both effective
and directing packets to the intended service functions, service path
tracing, diagnostic and fault isolation, alarm reporting, performance
measurement, locking and testing of service functions, validation
with the control plane (see Section 5.2), and also allow for vendor-
specific as well as experimental functions. SFC should leverage and,
if needed, extend relevant existing OAM mechanisms.
5.8. Resilience and Redundancy
As a practical operational requirement, any service chaining solution
needs to be able to respond effectively, and usually very quickly, to
failure conditions. These may be failures of connectivity in the
network between SFFs, failures of SFFs, or failures of SFs. Per-SF
state (as, for example, stateful-firewall state) is the
responsibility of the SF, and not addressed by this architecture.
Multiple techniques are available to address this issue. Solutions
can describe both what they require and what they allow to address
failure. Solutions can make use of flexible specificity of service
function paths, if the SFF can be given enough information in a
timely fashion to do this. Solutions can also make use of MAC- or
IP-level redundancy mechanisms such as Virtual Router Redundancy
Protocol (VRRP). Also, particularly for SF failures, load balancers
co-located with the SFF or as part of the service function delivery
mechanism can provide such robustness.
Similarly, operational requirements imply resilience in the face of
load changes. While mechanisms for managing (e.g., monitoring,
instantiating, loading images, providing configuration to SFC
control, deleting, etc.) virtual machines are out of scope for this
architecture, solutions can and are aided by describing how they can
make use of scaling mechanisms.
6. Security Considerations
The architecture described here is different from the current model,
and moving to the new model could lead to different security
arrangements and modeling. In the SFC architecture, a relatively
static topologically-dependent deployment model is replaced with the
chaining of sets of service functions. This can change the flow of
data through the network, and the security and privacy considerations
of the protocol and deployment will need to be reevaluated in light
of the new model.
Security considerations apply to the realization of this
architecture, in particular to the documents that will define
protocols. Such realization ought to provide means to protect
against security and privacy attacks in the areas hereby described.
Building from the categorization of [RFC7498], we can largely divide
the security considerations into four areas:
Service Overlay: Underneath the service function forwarders, the
components that are responsible for performing the transport
forwarding consult the outer-transport encapsulation for
underlay forwarding. Used transport mechanisms should satisfy
the security requirements of the specific SFC deployment. These
requirements typically include varying degrees of traffic
separation, protection against different attacks (e.g.,
spoofing, man-in-the-middle, brute-force, or insertion attacks),
and can also include authenticity and integrity checking, and/or
confidentiality provisions, for both the network overlay
transport and traffic it encapsulates.
Boundaries: Specific requirements may need to be enforced at the
boundaries of an SFC-enabled domain. These include, for
example, to avoid leaking SFC information, and to protect its
borders against various forms of attacks. If untrusted parties
can inject packets that will be treated as being properly
classified for service chaining, there are a large range of
attacks that can be mounted against the resulting system.
Depending upon deployment details, these likely include spoofing
packets from users and creating DDoS and reflection attacks of
various kinds. Thus, when transport mechanisms are selected for
use with SFC, they MUST ensure that outside parties cannot
inject SFC packets that will be accepted for processing into the
domain. This border security MUST include any tunnels to other
domains. If those tunnels are to be used for SFC without
reclassification, then the tunnel MUST include additional
techniques to ensure the integrity and validity of such packets.
Classification: Classification is used at the ingress edge of an
SFC-enabled domain. Policy for this classification is done
using a plurality of methods. Whatever method is used needs to
consider a range of security issues. These include appropriate
authentication and authorization of classification policy,
potential confidentiality issues of that policy, protection
against corruption, and proper application of policy with needed
segregation of application. This includes proper controls on
the policies that drive the application of the SFC encapsulation
and associated metadata to packets. Similar issues need to be
addressed if classification is performed within a service
chaining domain, i.e., reclassification.
SFC Encapsulation: The SFC encapsulation provides at a minimum SFP
identification, and carries metadata. An operator may consider
the SFC Metadata as sensitive. From a privacy perspective, a
user may be concerned about the operator revealing data about
(and not belonging to) the customer. Therefore, solutions
should consider whether there is a risk of sensitive information
slipping out of the operator's control. Issues of information
exposure should also consider flow analysis. Further, when a
specific metadata element is defined, it should be carefully
considered whether origin authentication is needed for it.
A classifier may have privileged access to information about a
packet or inside a packet (see Section 3, item 4, and
Section 4.9) that is then communicated in the metadata. The
threat of leaking this private data needs to be mitigated
[RFC6973]. As one example, if private data is represented by an
identifier, then a new identifier can be allocated, such that
the mapping from the private data to the new identifier is not
Some metadata added to and carried in SFC packets is sensitive
for various reasons, including potentially revealing personally
identifying information. Realizations of the architecture MUST
protect such information to ensure that it is handled with
suitable care and precautions against inappropriate
dissemination. This can have implications to the data plane,
the control plane, or both. Data-plane protocol definitions for
SFC can include suitable provisions to protect such information
for use when handling sensitive information, with packet or SFP
granularity. Equally, the control mechanisms used with SFC can
have provisions to determine that such mechanisms are available,
and to ensure that they are used when needed. Inability to do
so needs to result in error indications to appropriate
management systems. In particular, when the control systems
know that sensitive information may potentially be added to
packets at certain points on certain service chains, the control
mechanism MUST verify that appropriate protective treatment of
NSH information is available from the point where the
information is added to the point where it will be removed. If
such mechanisms are unavailable, error notifications SHOULD be
Additionally, SFC OAM functions need to not negatively affect the
security considerations of an SFC-enabled domain.
Finally, all entities (software or hardware) interacting with the
service chaining mechanisms need to provide means of security against
malformed, poorly configured (deliberate or not) protocol constructs
and loops. These considerations are largely the same as those in any
network, particularly an overlay network.
7.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
7.2. Informative References
Boucadair, M., Jacquenet, C., Parker, R., Lopez, D.,
Guichard, J., and C. Pignataro, "Service Function
Chaining: Framework & Architecture", Work in Progress,
draft-boucadair-sfc-framework-02, February 2014.
Quinn, P. and J. Halpern, "Service Function Chaining (SFC)
Architecture", Work in Progress, draft-quinn-sfc-arch-05,
[RFC3022] Srisuresh, P. and K. Egevang, "Traditional IP Network
Address Translator (Traditional NAT)", RFC 3022,
DOI 10.17487/RFC3022, January 2001,
[RFC6146] Bagnulo, M., Matthews, P., and I. van Beijnum, "Stateful
NAT64: Network Address and Protocol Translation from IPv6
Clients to IPv4 Servers", RFC 6146, DOI 10.17487/RFC6146,
April 2011, <http://www.rfc-editor.org/info/rfc6146>.
[RFC6296] Wasserman, M. and F. Baker, "IPv6-to-IPv6 Network Prefix
Translation", RFC 6296, DOI 10.17487/RFC6296, June 2011,
[RFC6973] Cooper, A., Tschofenig, H., Aboba, B., Peterson, J.,
Morris, J., Hansen, M., and R. Smith, "Privacy
Considerations for Internet Protocols", RFC 6973,
DOI 10.17487/RFC6973, July 2013,
[RFC7498] Quinn, P., Ed. and T. Nadeau, Ed., "Problem Statement for
Service Function Chaining", RFC 7498,
DOI 10.17487/RFC7498, April 2015,
The editors would like to thank Sam Aldrin, Alia Atlas, Nicolas
Bouthors, Stewart Bryant, Linda Dunbar, Alla Goldner, Ken Gray, Barry
Greene, Anil Gunturu, David Harrington, Shunsuke Homma, Dave Hood,
Chris Inacio, Nagendra Kumar, Hongyu Li, Andrew Malis, Guy
Meador III, Kengo Naito, Thomas Narten, Ron Parker, Reinaldo Penno,
Naiming Shen, Xiaohu Xu, and Lucy Yong for a thorough review and
The initial draft of this document was the result of merging two
previous documents, and this section lists the acknowledgments from
From "Service Function Chaining (SFC) Architecture" [Quinn2014]
The authors would like to thank David Ward, Abhijit Patra, Nagaraj
Bagepalli, Darrel Lewis, Ron Parker, Lucy Yong, and Christian
Jacquenet for their review and comments.
From "Service Function Chaining (SF) - Framework and Architecture"
Many thanks to D. Abgrall, D. Minodier, Y. Le Goff, D. Cheng,
R. White, and B. Chatras for their review and comments.
As noted above, this document is the result of merging two previous
documents. This section lists those who provided important ideas and
text that fed into this architecture.
The authors of "Service Function Chaining (SFC) - Framework and
Architecture" [Boucadair2014] were:
Diego R. Lopez
The contributors were:
The authors of "Service Function Chaining (SFC) Architecture"
Paul Quinn (editor)
Joel Halpern (editor)
The contributors were:
Joel Halpern (editor)
Carlos Pignataro (editor)
Cisco Systems, Inc.