RFC 8459






Internet Engineering Task Force (IETF)                         D. Dolson
Request for Comments: 8459                                      Sandvine
Category: Experimental                                          S. Homma
ISSN: 2070-1721                                                      NTT
                                                                D. Lopez
                                                          Telefonica I+D
                                                            M. Boucadair
                                                                  Orange
                                                          September 2018


             Hierarchical Service Function Chaining (hSFC)

Abstract



   Hierarchical Service Function Chaining (hSFC) is a network
   architecture allowing an organization to decompose a large-scale
   network into multiple domains of administration.

   The goals of hSFC are to make a large-scale network easier to design,
   simpler to control, and supportive of independent functional groups
   within large network operators.

Status of This Memo



   This document is not an Internet Standards Track specification; it is
   published for examination, experimental implementation, and
   evaluation.

   This document defines an Experimental Protocol for the Internet
   community.  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 candidates for any level of
   Internet Standard; see Section 2 of RFC 7841.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   https://www.rfc-editor.org/info/rfc8459.











Dolson, et al.                Experimental                      [Page 1]

RFC 8459                          hSFC                    September 2018


Copyright Notice



   Copyright (c) 2018 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
   (https://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.





































Dolson, et al.                Experimental                      [Page 2]

RFC 8459                          hSFC                    September 2018


Table of Contents



   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   4
     1.1.  Experiment Goals  . . . . . . . . . . . . . . . . . . . .   5
   2.  Terminology . . . . . . . . . . . . . . . . . . . . . . . . .   6
   3.  Hierarchical Service Function Chaining (hSFC) . . . . . . . .   6
     3.1.  Upper Level . . . . . . . . . . . . . . . . . . . . . . .   6
     3.2.  Lower Levels  . . . . . . . . . . . . . . . . . . . . . .   8
   4.  Internal Boundary Node (IBN)  . . . . . . . . . . . . . . . .  10
     4.1.  IBN Path Configuration  . . . . . . . . . . . . . . . . .  10
       4.1.1.  Flow-Stateful IBN . . . . . . . . . . . . . . . . . .  11
       4.1.2.  Encoding Upper-Level Paths in Metadata  . . . . . . .  12
       4.1.3.  Using Unique Paths per Upper-Level Path . . . . . . .  13
       4.1.4.  Nesting Upper-Level NSH within Lower-Level NSH  . . .  13
       4.1.5.  Stateful/Metadata Hybrid  . . . . . . . . . . . . . .  14
     4.2.  Gluing Levels Together  . . . . . . . . . . . . . . . . .  16
     4.3.  Decrementing Service Index  . . . . . . . . . . . . . . .  16
     4.4.  Managing TTL  . . . . . . . . . . . . . . . . . . . . . .  16
   5.  Subdomain Classifier  . . . . . . . . . . . . . . . . . . . .  17
   6.  Control Plane Elements  . . . . . . . . . . . . . . . . . . .  18
   7.  Extension for Adapting to NSH-Unaware Service Functions . . .  18
     7.1.  Purpose . . . . . . . . . . . . . . . . . . . . . . . . .  19
     7.2.  Requirements for an IBN . . . . . . . . . . . . . . . . .  20
   8.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .  21
   9.  Security Considerations . . . . . . . . . . . . . . . . . . .  21
     9.1.  Control Plane . . . . . . . . . . . . . . . . . . . . . .  21
     9.2.  Infinite Forwarding Loops . . . . . . . . . . . . . . . .  22
   10. References  . . . . . . . . . . . . . . . . . . . . . . . . .  22
     10.1.  Normative References . . . . . . . . . . . . . . . . . .  22
     10.2.  Informative References . . . . . . . . . . . . . . . . .  22
   Appendix A.  Examples of Hierarchical Service Function Chaining .  24
     A.1.  Reducing the Number of Service Function Paths . . . . . .  24
     A.2.  Managing a Distributed DC Network . . . . . . . . . . . .  26
   Acknowledgements  . . . . . . . . . . . . . . . . . . . . . . . .  28
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .  29
















Dolson, et al.                Experimental                      [Page 3]

RFC 8459                          hSFC                    September 2018


1.  Introduction



   Service Function Chaining (SFC) is a technique for prescribing
   differentiated traffic-forwarding policies within an SFC-enabled
   domain.  The SFC architecture is described in detail in [RFC7665] and
   is not repeated here.

   This document focuses on the difficult problem of implementing SFC
   across a large, geographically dispersed network, potentially
   comprised of millions of hosts and thousands of network-forwarding
   elements and which may involve multiple operational teams (with
   varying functional responsibilities).  We recognize that some
   stateful Service Functions (SFs) require bidirectional traffic for
   transport-layer sessions (e.g., NATs, firewalls).  We assume that
   some Service Function Paths (SFPs) need to be selected on the basis
   of transport-layer coordinate (typically, the 5-tuple of source IP
   address, source port number, destination IP address, destination port
   number, and transport protocol) stickiness to specific stateful SF
   instances.

   Difficult problems are often made easier by decomposing them in a
   hierarchical (nested) manner.  So, instead of considering a single
   SFC control plane that can manage (create, withdraw, supervise, etc.)
   complete SFPs from one end of the network to the other, we decompose
   the network into smaller domains operated by as many SFC control
   plane components (under the same administrative entity).
   Coordination between such components is further discussed in this
   document.

   Each subdomain may support a subset of the network applications or a
   subset of the users.  Decomposing a network should be done with care
   to ease monitoring and troubleshooting of the network and services as
   a whole.  The criteria for decomposing a domain into multiple SFC-
   enabled subdomains are beyond the scope of this document.  These
   criteria are deployment specific.

   An example of simplifying a network by using multiple SFC-enabled
   domains is further discussed in [USE-CASES].

   We assume the SFC-aware nodes use the Network Service Header (NSH)
   [RFC8300] or a similar labeling mechanism.  Examples are described in
   Appendix A.

   The SFC-enabled domains discussed in this document are assumed to be
   under the control of a single organization (an operator, typically),
   such that there is a strong trust relationship between the domains.
   The intention of creating multiple domains is to improve the ability




Dolson, et al.                Experimental                      [Page 4]

RFC 8459                          hSFC                    September 2018


   to operate a network.  It is outside of the scope of this document to
   consider domains operated by different organizations or dwell on
   interoperator considerations.

   We introduce the concept of an Internal Boundary Node (IBN) that acts
   as a gateway between the levels of the hierarchy.  We also discuss
   options for realizing this function.

1.1.  Experiment Goals



   This document defines an architecture that aims to solve
   complications that may be encountered when deploying SFC in large
   networks.  A single network is therefore decomposed into multiple
   subdomains, each treated as an SFC-enabled domain.  Levels of
   hierarchy are defined, together with SFC operations that are specific
   to each level.  In order to ensure consistent SFC operations when
   multiple subdomains are involved, this document identifies and
   analyzes various options for IBNs to glue the layers together
   (Section 4.1).

   Because it does not make any assumptions about (1) how subdomains are
   defined, (2) whether one or multiple IBNs are enabled per subdomain,
   (3) whether the same IBN is solicited at both the ingress and egress
   of a subdomain for the same flow, (4) the nature of the internal
   paths to reach SFs within a subdomain, or (5) the lack of deployment
   feedback, this document does not call for a recommended option to
   glue the SFC layers together.

   Further experiments are required to test and evaluate the different
   options.  A recommendation for hSFC might be documented in a future
   specification when the results of implementation and deployment of
   the aforementioned options are available.

   It is not expected that all the options discussed in this document
   will be implemented and deployed.  The lack of an implementation
   might be seen as a signal to recommend against a given option.















Dolson, et al.                Experimental                      [Page 5]

RFC 8459                          hSFC                    September 2018


2.  Terminology



   This document makes use of the terms defined in Section 1.4 of
   [RFC7665] and Section 1.3 of [RFC8300].

   The following terms are defined:

   o  Upper-level domain: the entire network domain to be managed.

   o  Lower-level domain: a portion of the network (called a subdomain).

   o  Internal Boundary Node (IBN): is responsible for bridging packets
      between upper and lower levels of SFC-enabled domains.

3.  Hierarchical Service Function Chaining (hSFC)



   A hierarchy has multiple levels: the topmost level encompasses the
   entire network domain to be managed, and lower levels encompass
   portions of the network.  These levels are discussed in the following
   subsections.

3.1.  Upper Level



   Considering the example depicted in Figure 1, a top-level network
   domain includes SFC data plane components distributed over a wide
   area, including:

   o  Classifiers (CFs)

   o  Service Function Forwarders (SFFs)

   o  Subdomains



















Dolson, et al.                Experimental                      [Page 6]

RFC 8459                          hSFC                    September 2018


                    +------------+
                    |Subdomain#1 |
                    |  in DC1    |
                    +----+-------+
                         |
                 .---- SFF1 ------.   +----+
       +----+   /     /  |         \--|CF#4|
   --->|CF#1|--/---->'   |          \ +----+
       +----+ /  SC#1    |           \
              |          |            |
              |          V    .------>|--->
              |         /    /        |
               \         |   /        /
        +----+  \        |  /        /  +----+
        |CF#2|---\       | /        /---|CF#3|
        +----+    '---- SFF2 ------'    +----+
                         |
                    +----+-------+
                    |Subdomain#2 |
                    |   in DC2   |
                    +------------+

       Legend:
         SC#1: Service Chain 1
           DC: Data Center

     Figure 1: Network-Wide View of Upper Level of Hierarchy

   One path is shown from edge classifier (CF#1) to SFF1 to Subdomain#1
   (residing in Data Center 1) to SFF1 to SFF2 (residing in Data Center
   2) to Subdomain#2 to SFF2 to network egress.

   For the sake of clarity, components of the underlay network are not
   shown; an underlay network is assumed to provide connectivity between
   SFC data plane components.

   Top-level SFPs carry packets from classifiers through a set of SFFs
   and subdomains, with the operations within subdomains being opaque to
   the upper levels.

   We expect the system to include a top-level control plane having
   responsibility for configuring forwarding policies and traffic-
   classification rules.

   The top-level Service Chaining control plane manages end-to-end
   service chains and associated service function paths from network
   edge points to subdomains.  It also configures top-level classifiers




Dolson, et al.                Experimental                      [Page 7]

RFC 8459                          hSFC                    September 2018


   at a coarse level (e.g., based on source or destination host) to
   forward traffic along paths that will transit across appropriate
   subdomains.

   Figure 1 shows one possible service chain passing from the edge
   through two subdomains to network egress.  The top-level control
   plane does not configure traffic-classification rules or forwarding
   policies within the subdomains.

   At this network-wide level, the number of SFPs required is a linear
   function of the number of ways in which a packet is required to
   traverse different subdomains and egress the network.  Note that the
   various paths that may be followed within a subdomain are not
   represented by distinct network-wide SFPs; specific policies at the
   ingress nodes of each subdomain bind flows to subdomain paths.

   Packets are classified at the edge of the network to select the paths
   by which subdomains are to be traversed.  At the ingress of each
   subdomain, packets are reclassified to paths directing them to the
   required SFs of the subdomain.  At the egress of each subdomain,
   packets are returned to the top-level paths.  Contrast this with an
   approach requiring the top-level classifier to select paths to
   specify all of the SFs in each subdomain.

   It should be assumed that some SFs require bidirectional symmetry of
   paths (see more in Section 5).  Therefore, the classifiers at the top
   level must be configured with policies ensuring outgoing packets take
   the reverse path of incoming packets through subdomains.

3.2.  Lower Levels



   Each of the subdomains in Figure 1 is an SFC-enabled domain.

   Figure 2 shows a subdomain interfaced with an upper-level domain by
   means of an Internal Boundary Node (IBN).  An IBN acts as an SFC-
   aware SF in the upper-level domain and as a classifier in the lower-
   level domain.  As such, data packets entering the subdomain are
   already SFC encapsulated.  Also, it is the purpose of the IBN to
   apply classification rules and direct the packets to the selected
   local SFPs terminating at an egress IBN.  Finally, the egress IBN
   restores packets to the original SFC shim and hands them off to SFFs.

   Each subdomain intersects a subset of the total paths that are
   possible in the upper-level domain.  An IBN is concerned with upper-
   level paths, but only those traversing its subdomain.






Dolson, et al.                Experimental                      [Page 8]

RFC 8459                          hSFC                    September 2018


   Each subdomain is likely to have a control plane that can operate
   independently of the top-level control plane, managing
   classification, forwarding paths, etc., within the level of the
   subdomain, with the details being opaque to the upper-level control
   elements.  Section 4 provides more details about the behavior of an
   IBN.

   The subdomain control plane configures the classification rules in
   the IBN, where SFC encapsulation of the top-level domain is converted
   to/from SFC encapsulation of the lower-level domain.  The subdomain
   control plane also configures the forwarding rules in the SFFs of the
   subdomain.

     +----+    +-----+  +----------------------+   +-----+
     |    |    | SFF |  |   IBN 1  (in DC 1)   |   | SFF |
     |    |SC#1|     |  |  +----------------+  |   |     |
   ->|    |===============>|      SFF       |================>
     |    |    +-----+  |  +----------------+  |   +-----+
     | CF |             |   |              ^   |
     |    |             |   v              |   |
     |    |             |+--------------------+|   Upper domain
     |    |             ||CF, fwd/rev mapping ||
     |    |    * * * * *||  and "glue"        || * * * * *
     |    |    *        |+--------------------+|         *
     +----+    *        | | |              | | |    Sub  *
               *        +-o-o--------------o-o-+   domain*
               *     SC#2 | |SC#1          ^ ^       #1  *
               *    +-----+ |              | |           *
               *    |       V              | |           *
               *    |     +---+  +------+  | |           *
               *    |     |SFF|->|SF#1.1|--+ |           *
               *    |     +---+  +------+    |           *
               *    V                        |           *
               *  +---+  +------+  +---+  +------+       *
               *  |SFF|->|SF#2.1|->|SFF|->|SF#2.2|       *
               *  +---+  +------+  +---+  +------+       *
               * * * * * * * * * * * * * * * * * * * * * *
   Legend:
        *** Subdomain boundary
        === upper-level chain
        --- lower-level chain

       Figure 2: Example of a Subdomain within an Upper-Level Domain

   If desired, the pattern can be applied recursively.  For example,
   SF#1.1 in Figure 2 could be a subdomain of the subdomain.





Dolson, et al.                Experimental                      [Page 9]

RFC 8459                          hSFC                    September 2018


4.  Internal Boundary Node (IBN)



   As mentioned in the previous section, a network element termed an
   "Internal Boundary Node" (or IBN) is responsible for bridging packets
   between upper and lower layers of SFC-enabled domains.  It behaves as
   an SF to the upper level (Section 3.1) and looks like a classifier
   and end of chain to the lower level (Section 3.2).

   To achieve the benefits of hierarchy, the IBN should be applying
   fine-grained traffic-classification rules at a lower level than the
   traffic passed to it.  This means that the number of SFPs within the
   lower level is greater than the number of SFPs arriving to the IBN.

   The IBN is also the termination of lower-level SFPs.  This is because
   the packets exiting lower-level SFPs must be returned to the upper-
   level SFPs and forwarded to the next hop in the upper-level domain.

   When different metadata schemes are used at different levels, the IBN
   has further responsibilities: when packets enter the subdomain, the
   IBN translates upper-level metadata into lower-level metadata; and
   when packets leave the subdomain at the termination of lower-level
   SFPs, the IBN translates lower-level metadata into upper-level
   metadata.

   Appropriately configuring IBNs is key to ensuring the consistency of
   the overall SFC operation within a given domain that enables hSFC.
   Classification rules (or lack thereof) in the IBN classifier can, of
   course, impact upper levels.

4.1.  IBN Path Configuration



   The lower-level domain may be provisioned with valid upper-level
   paths or allow any upper-level paths.

   When packets enter the subdomain, the Service Path Identifier (SPI)
   and Service Index (SI) are re-marked according to the path selected
   by the (subdomain) classifier.

   At the termination of an SFP in the subdomain, packets can be
   restored to an original upper-level SFP by implementing one of these
   methods:

   1.  Saving the SPI and SI in transport-layer flow state
       (Section 4.1.1).

   2.  Pushing the SPI and SI into a metadata header (Section 4.1.2).





Dolson, et al.                Experimental                     [Page 10]

RFC 8459                          hSFC                    September 2018


   3.  Using unique lower-level paths per upper-level path coordinates
       (Section 4.1.3).

   4.  Nesting NSH headers, encapsulating the upper-level NSH headers
       within the lower-level NSH headers (Section 4.1.4).

   5.  Saving the upper level with a flow identifier (ID) and placing an
       hSFC Flow ID into a metadata header (Section 4.1.5).



4.1.1.  Flow-Stateful IBN



   An IBN can be flow aware, returning packets to the correct upper-
   level SFP on the basis, for example, of the transport-layer
   coordinates (typically, a 5-tuple) of packets exiting the lower-level
   SFPs.

   When packets are received by the IBN on an upper-level path, the
   classifier parses encapsulated packets for IP and transport-layer
   (TCP, UDP, etc.) coordinates.  State is created, indexed by some or
   all transport coordinates (typically, {source-IP, destination-IP,
   source-port, destination-port, and transport protocol}).  The state
   contains, at minimum, the critical fields of the encapsulating SFC
   header (SPI, SI, MD Type, flags); additional information carried in
   the packet (metadata, TTL) may also be extracted and saved as state.
   Note that some fields of a packet may be altered by an SF of the
   subdomain (e.g., source IP address).

   Note that this state is only accessed by the classifier and
   terminator functions of the subdomain.  Neither the SFFs nor SFs have
   knowledge of this state; in fact they may be agnostic about being in
   a subdomain.

   One approach is to ensure that packets are terminated at the end of
   the chain at the same IBN that classified the packet at the start of
   the chain.  If the packet is returned to a different egress IBN,
   state must be synchronized between the IBNs.

   When a packet returns to the IBN at the end of a chain (which is the
   SFP-terminating node of the lower-level chain), the SFC header is
   removed, the packet is parsed for flow-identifying information, and
   state is retrieved from within the IBN using the flow-identifying
   information as index.

   State cannot be created by packets arriving from the lower-level
   chain; when state cannot be found for such packets, they must be
   dropped.





Dolson, et al.                Experimental                     [Page 11]

RFC 8459                          hSFC                    September 2018


   This stateful approach is limited to use with SFs that retain the
   transport coordinates of the packet.  This approach cannot be used
   with SFs that modify those coordinates (e.g., NATs) or otherwise
   create packets for new coordinates other than those received (e.g.,
   as an HTTP cache might do to retrieve content on behalf of the
   original flow).  In both cases, the fundamental problem is the
   inability to forward packets when state cannot be found for the
   packet transport-layer coordinates.

   In the stateful approach, there are issues caused by having state,
   such as how long the state should be maintained as well as whether
   the state needs to be replicated to other devices to create a highly
   available network.

   It is valid to consider the state to be disposable after failure,
   since it can be recreated by each new packet arriving from the upper-
   level domain.  For example, if an IBN loses all flow state, the state
   is recreated by an endpoint retransmitting a TCP packet.

   If an SFC domain handles multiple network regions (e.g., multiple
   private networks), the coordinates may be augmented with additional
   parameters, perhaps using some metadata to identify the network
   region.

   In this stateful approach, it is not necessary for the subdomain's
   control plane to modify paths when upper-level paths are changed.
   The complexity of the upper-level domain does not cause complexity in
   the lower-level domain.

   Since it doesn't depend on NSH in the lower-level domain, this flow-
   stateful approach can be applied to translation methods of converting
   NSH to other forwarding techniques (refer to Section 7).

4.1.2.  Encoding Upper-Level Paths in Metadata



   An IBN can push the upper-level SPI and SI (or encoding thereof) into
   a metadata field of the lower-level encapsulation (e.g., placing
   upper-level path information into a metadata field of the NSH).  When
   packets exit the lower-level path, the upper-level SPI and SI can be
   restored from the metadata retrieved from the packet.

   This approach requires the SFs in the path to be capable of
   forwarding the metadata and appropriately attaching metadata to any
   packets injected for a flow.

   Using a new metadata header may inflate packet size when variable-
   length metadata (NSH MD Type 0x2) is used.




Dolson, et al.                Experimental                     [Page 12]

RFC 8459                          hSFC                    September 2018


   It is conceivable that the MD Type 0x1 Fixed-Length Context Header
   field of the NSH is not all relevant to the lower-level domain.  In
   this case, 32 bits of the Fixed-Length Context Header field could be
   repurposed within the lower-level domain and restored when leaving.

   If flags or TTL (see Section 4.4) from the original header also need
   to be saved, more metadata space will be consumed.

   In this metadata approach, it is not necessary for the subdomain's
   control element to modify paths when upper-level paths are changed.
   The complexity of the upper-level domain does not increase complexity
   in the lower-level domain.

4.1.3.  Using Unique Paths per Upper-Level Path



   This approach assumes that paths within the subdomain are constrained
   so that an SPI (of the subdomain) unambiguously indicates the egress
   SPI and SI (of the upper domain).  This allows the original path
   information to be restored at subdomain egress from a look-up table
   using the subdomain SPI.

   Whenever the upper-level domain provisions a path via the lower-level
   domain, the lower-level domain control plane must provision
   corresponding paths to traverse the lower-level domain.

   A downside of this approach is that the number of paths in the lower-
   level domain is multiplied by the number of paths in the upper-level
   domain that traverse the lower-level domain.  That is, a subpath must
   be created for each combination of upper SPI/SI and lower chain.  The
   number of paths required for lower-level domains will increase
   exponentially as hierarchy becomes deep.

   A further downside of this approach is that it requires upper and
   lower levels to utilize the same metadata configuration.

   Furthermore, this approach does not allow any information to be
   stashed away in state or embedded in metadata.  For example, the TTL
   modifications by the lower level cannot be hidden from the upper
   level.

4.1.4.  Nesting Upper-Level NSH within Lower-Level NSH



   When packets arrive at an IBN in the top-level domain, the classifier
   in the IBN determines the path for the lower-level domain and pushes
   the new NSH header in front of the original NSH header.






Dolson, et al.                Experimental                     [Page 13]

RFC 8459                          hSFC                    September 2018


   As shown in Figure 3, the Lower-NSH header used to forward packets in
   the lower-level domain precedes the Upper-NSH header from the top-
   level domain.

                    +---------------------------------+
                    |  Outer-Transport Encapsulation  |
                    +---------------------------------+
                    |        Lower-NSH Header         |
                    +---------------------------------+
                    |        Upper-NSH Header         |
                    +---------------------------------+
                    |          Original Packet        |
                    +---------------------------------+

                 Figure 3: Encapsulation of NSH within NSH

   The traffic with this stack of two NSH headers is to be forwarded
   according to the Lower-NSH header in the lower-level SFC domain.  The
   Upper-NSH header is preserved in the packets but not used for
   forwarding.  At the last SFF of the chain of the lower-level domain
   (which resides in the IBN), the Lower-NSH header is removed from the
   packet, and then the packet is forwarded by the IBN to an SFF of the
   upper-level domain.  The packet will be forwarded in the top-level
   domain according to the Upper-NSH header.

   With such encapsulation, Upper-NSH information is carried along the
   extent of the lower-level chain without modification.

   A benefit of this approach is that it does not require state in the
   IBN or configuration to encode fields in metadata.  All header
   fields, including flags and TTL, are easily restored when the chains
   of the subdomain terminate.

   However, the downside is that it does require SFC-aware SFs in the
   lower-level domain to be able to parse multiple NSH layers.  If an
   SFC-aware SF injects packets, it must also be able to deal with
   adding appropriate multiple layers of headers to injected packets.

   By increasing packet overhead, nesting may lead to fragmentation or
   decreased MTU in some networks.

4.1.5.  Stateful/Metadata Hybrid



   The basic idea of this approach is for the IBN to save upper domain
   encapsulation information such that it can be retrieved by a unique
   identifier, termed an "hSFC Flow ID".





Dolson, et al.                Experimental                     [Page 14]

RFC 8459                          hSFC                    September 2018


   The hSFC Flow ID is placed, for example, in the NSH Fixed-Length
   Context Header field of the packet in the lower-level domain, as
   shown in Figure 4.  Likewise, hSFC Flow ID may be encoded as a
   Variable-Length Context Header field when MD Type 0x2 is used.

   When packets exit the lower-level domain, the IBN uses the hSFC Flow
   ID to retrieve the appropriate NSH encapsulation for returning the
   packet to the upper domain.  The hSFC Flow ID Context Header is then
   stripped by the IBN.

      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |Ver|O|U|    TTL    |   Length  |U|U|U|U|MD Type| Next Protocol |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |          Service Path Identifier              | Service Index |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
     |                      hSFC Flow ID                             |
     |              Zero Padding or other fields                     |
     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

             Figure 4: Storing hSFC Flow ID in Lower-Level NSH
        Fixed-Length Context Header Field ([RFC8300], Section 2.4)

   Advantages of this approach include:

   o  It does not require state to be based on a 5-tuple, so it works
      with SFs that change the IP addresses or port numbers of a packet,
      such as NATs.

   o  It does not require all domains to have the same metadata scheme.

   o  It can be used to restore any upper-domain information, including
      metadata, flags, and TTL, not just the service path.

   o  The lower-level domain only requires a single item of metadata
      regardless of the number of items of metadata used in the upper
      domain.

   o  The SFC-related functionality required by this approach in an SFC-
      aware SF is able to preserve and apply metadata, which is a
      requirement that was already present in [RFC8300].

   Disadvantages include those of other stateful approaches, including
   state timeout and synchronization, mentioned in Section 4.1.1.







Dolson, et al.                Experimental                     [Page 15]

RFC 8459                          hSFC                    September 2018


   There may be a large number of unique NSH encapsulations to be
   stored, given that the hSFC Flow ID must represent all of the bits in
   the upper-level encapsulation.  This might consume a lot of memory or

   create out-of-memory situations in which hSFC Flow IDs cannot be
   created or old hSFC Flow IDs are discarded while still in use.

4.2.  Gluing Levels Together



   The SPI or metadata included in a packet received by the IBN may be
   used as input to reclassification and path selection within a lower-
   level domain.

   In some cases, the meanings of the various path IDs and metadata must
   be coordinated between domains for the sake of proper end-to-end SFC
   operation.

   One approach is to use well-known identifier values in metadata,
   maintained in a global registry.

   Another approach is to use well-known labels for chain identifiers or
   metadata, as an indirection to the actual identifiers.  The actual
   identifiers can be assigned by control-plane systems.  For example, a
   subdomain classifier could have a policy, "if pathID = classA then
   chain packet to path 1234"; the upper-level controller would be
   expected to configure the concrete upper-level "pathID" for "classA".

4.3.  Decrementing Service Index



   Because the IBN acts as an SFC-aware SF to the upper-level domain, it
   must decrement the Service Index in the NSH headers of the upper-
   level path.  This operation should be undertaken when the packet is
   first received by the IBN, before applying any of the strategies of
   Section 4.1, immediately prior to classification.

4.4.  Managing TTL



   The NSH base header contains a TTL field [RFC8300].  There is a
   choice:

      a subdomain may appear as a pure service function, which should
      not decrement the TTL from the perspective of the upper-level
      domain, or

      all of the TTL changes within the subdomain may be visible to the
      upper-level domain.





Dolson, et al.                Experimental                     [Page 16]

RFC 8459                          hSFC                    September 2018


   Some readers may recognize this as a choice between "pipe" and
   "uniform" models, respectively [RFC3443].

   The network operator should be given control of this behavior,
   choosing whether to expose the lower-level topology to the upper
   layer.  An implementation may support per-packet policy, allowing
   some users to perform a layer-transcending trace route, for example.

   The choice affects whether the methods of restoring the paths in
   Section 4.1 restore a saved version of the TTL or propagate it with
   the packet.  The method of Section 4.1.3 does not permit topology
   hiding.  The other methods of Sections 4.1.1, 4.1.2, 4.1.4, and 4.1.5
   have unique methods for restoring saved versions of the TTL.

5.  Subdomain Classifier

   Within the subdomain (referring to Figure 2), as the classifier
   receives incoming packets, the upper-level encapsulation is treated
   according to one of the methods described in Section 4.1 to either
   statefully store, encode, or nest header information.  The classifier
   then selects the path and metadata for the packet within the
   subdomain.

   One of the goals of the hierarchical approach is to make it easy to
   have transport-flow-aware service chaining with bidirectional paths.
   For example, it is desired that for each TCP flow, the client-to-
   server packets traverse the same SF instances as the server-to-client
   packets, but in the opposite sequence.  We call this "bidirectional
   symmetry".  If bidirectional symmetry is required, it is the
   responsibility of the control plane to be aware of symmetric paths
   and configure the classifier to chain the traffic in a symmetric
   manner.

   Another goal of the hierarchical approach is to simplify the
   mechanisms of scaling SFs in and out.  All of the complexities of
   load-balancing among multiple SFs can be handled within a subdomain,
   under control of the classifier, allowing the upper-level domain to
   be oblivious to the existence of multiple SF instances.

   Considering the requirements of bidirectional symmetry and load-
   balancing, it is useful to have all packets entering a subdomain be
   received by the same classifier or a coordinated cluster of
   classifiers.  There are both stateful and stateless approaches to
   ensuring bidirectional symmetry.







Dolson, et al.                Experimental                     [Page 17]

RFC 8459                          hSFC                    September 2018


6.  Control Plane Elements



   Although SFC control protocols have not yet been standardized (as of
   2018), from the point of view of hierarchical service function
   chaining, we have these expectations:

   o  Each control-plane instance manages a single level of the
      hierarchy of a single domain.

   o  Each control plane is agnostic about other levels of the
      hierarchy.  This aspect allows humans to reason about the system
      within a single domain and control-plane algorithms to use only
      domain-local inputs.  Top-level control does not need visibility
      to subdomain policies, nor does subdomain control need visibility
      to upper-level policies.  (Top-level control considers a subdomain
      as though it were an SF.)

   o  Subdomain control planes are agnostic about the control planes of
      other subdomains.  This allows both humans and machines to
      manipulate subdomain policy without considering policies of other
      domains.

   Recall that the IBN acts as an SFC-aware SF in the upper-level domain
   (receiving SF instructions from the upper-level control plane) and as
   a classifier in the lower-level domain (receiving classification
   rules from the subdomain control plane).  In this view, it is the IBN
   that glues the layers together.

   These expectations are not intended to prohibit network-wide control.
   A control hierarchy can be envisaged to distribute information and
   instructions to multiple domains and subdomains.  Control hierarchy
   is outside the scope of this document.

7.  Extension for Adapting to NSH-Unaware Service Functions



   The hierarchical approach can be used for dividing networks into NSH-
   aware and NSH-unaware domains by converting NSH encapsulation to
   other forwarding techniques (e.g., 5-tuple-based forwarding with
   OpenFlow), as shown in Figure 5.












Dolson, et al.                Experimental                     [Page 18]

RFC 8459                          hSFC                    September 2018


                    * * * * * * * * * * * * * * * * * *
                  *   NSH-aware domain                 *
                  *       +-------+       +-------+    *
                  *       | SF#1  |       | SF#5  |    *
                  *       +-o---o-+       +-o---o-+    *
                  *         ^   |           ^   |      *
                  *       +-|---|-+       +-|---|-+    *
                  *       | |SFF| |       | |SFF| |    *
                  *       +-|---|-+       +-|---|-+    *
                  *         .   |           |   .      *
                  * +--+   /    |           |    \     *
                 -->|CF|--'     |           |     '------->
                  * +--+        v           |          *
                  *         +---o-----------o---+      *
                   .*.*.*.*.|  / |   IBN   | \  |*.*.*.
                  .         +-o--o---------o--o-+      .
                  .           |  |         ^  ^        .
                  .           |  +-+     +-+  |        .
                  .       +---+    v     |    +---+    .
                  .       |      +-o-----o-+      |    .
                  .       |      |  SF#2   |      |    .
                  .       |      +---------+      |    .
                  .       +--+                 +--+    .
                  .          |   +---------+   |       .
                  .          v   |         v   |       .
                  .        +-o---o-+     +-o---o-+     .
                  .        | SF#3  |     | SF#4  |     .
                  .        +-------+     +-------+     .
                  .   NSH-unaware domain               .
                   . . . . . . . . . . . . . . . . . .


   SF#1 and SF#5 are NSH aware; SF#2, SF#3, and SF#4 are NSH unaware.
   In the NSH-unaware domain, packets are conveyed in a format supported
   by SFs that are deployed there.

           Figure 5: Dividing NSH-Aware and NSH-Unaware Domains

7.1.  Purpose



   This approach is expected to facilitate service chaining in networks
   in which NSH-aware and NSH-unaware SFs coexist.  Some examples of
   such situations are:

   o  In a period of transition from legacy SFs to NSH-aware SFs

   o  Supporting multitenancy




Dolson, et al.                Experimental                     [Page 19]

RFC 8459                          hSFC                    September 2018


7.2.  Requirements for an IBN



   In this usage, an IBN classifier is required to have an NSH
   conversion table for applying packets to appropriate lower-level
   paths and returning packets to the correct upper-level paths.  For
   example, the following methods would be used for saving/restoring
   upper-level path information:

   o  Saving SPI and SI in transport-layer flow state (refer to
      Section 4.1.1)

   o  Using unique lower-level paths per upper-level NSH coordinates
      (refer to Section 4.1.3)

   Using the unique paths approach would be especially good for
   translating NSH to a different forwarding technique in the lower
   level.  A single path in the upper level may be branched to multiple
   paths in the lower level such that any lower-level path is only used
   by one upper-level path.  This allows unambiguous restoration to the
   upper-level path.

   In addition, an IBN might be required to convert metadata contained
   in the NSH to the format appropriate to the packet in the lower-level
   path.  For example, some legacy SFs identify subscribers based on
   information about the network topology, such as the VLAN ID (VID),
   and the IBN would be required to create a VLAN to packets from
   metadata if the subscriber identifier is conveyed as metadata in
   upper-level domains.

   Other fundamental functions required for an IBN (e.g., maintaining
   metadata of upper level or decrementing Service Index) are the same
   as in normal usage.

   It is useful to permit metadata to be transferred between levels of a
   hierarchy.  Metadata from an upper level may be useful within a
   subdomain, and a subdomain may augment metadata for consumption in an
   upper domain.  However, allowing uncontrolled metadata between
   domains may lead to forwarding failures.

      In order to prevent SFs of lower-level SFC-enabled domains from
      supplying (illegitimate) metadata, IBNs may be instructed to only
      permit specific metadata types to exit the subdomain.  Such
      control over the metadata in the upper level is the responsibility
      of the upper-level control plane.







Dolson, et al.                Experimental                     [Page 20]

RFC 8459                          hSFC                    September 2018


      To limit unintentional metadata reaching SFs of lower-level SFC-
      enabled subdomains, IBNs may be instructed to only permit specific
      metadata types into the subdomain.  Such control of metadata in
      the lower-level domain is the responsibility of the lower-level
      control plane.

8.  IANA Considerations



   This document has no IANA actions.

9.  Security Considerations



   hSFC makes use of service chaining architecture; hence, it inherits
   the security considerations described in the architecture document
   [RFC7665].

   Furthermore, hSFC inherits the security considerations of the data-
   plane protocols (e.g., NSH) and control-plane protocols used to
   realize the solution.

   This document describes systems that may be managed by distinct teams
   that all belong to the same administrative entity.  Subdomains must
   have consistent configurations in order to properly forward traffic.
   Any protocol designed to distribute the configurations must be secure
   from tampering.  The means of preventing attacks from within a
   network must be enforced.  For example, continuously monitoring the
   network may allow detecting such misbehaviors. hSFC adheres to the
   same security considerations as [RFC8300].  Those considerations must
   be taken into account.

   The options in Sections 4.1.2 and 4.1.5 assume the use of a dedicated
   context header to store information to bind a flow to its upper-level
   SFP.  Such a context header is stripped by the IBN of a subdomain
   before exiting a subdomain.  Additional guards to prevent leaking
   unwanted context information when entering/exiting a subdomain are
   discussed in Section 7.2.

   All of the systems and protocols must be secure from modification by
   untrusted agents.

9.1.  Control Plane



   Security considerations related to the control plane are discussed in
   the corresponding control specification documents (e.g.,
   [BGP-CONTROL], [PCEP-EXTENSIONS], or [RADIUS]).






Dolson, et al.                Experimental                     [Page 21]

RFC 8459                          hSFC                    September 2018


9.2.  Infinite Forwarding Loops



   Distributing policies among multiple domains may lead to forwarding
   loops.  NSH supports the ability to detect loops (as described in
   Sections 4.3 and 4.4 of [RFC8300]), but the means of ensuring the
   consistency of the policies should be enabled at all levels of a
   domain.  Within the context of hSFC, it is the responsibility of the
   Control Elements at all levels to prevent such (unwanted) loops.

10.  References



10.1.  Normative References



   [RFC7665]  Halpern, J., Ed. and C. Pignataro, Ed., "Service Function
              Chaining (SFC) Architecture", RFC 7665,
              DOI 10.17487/RFC7665, October 2015,
              <https://www.rfc-editor.org/info/rfc7665>.

   [RFC8300]  Quinn, P., Ed., Elzur, U., Ed., and C. Pignataro, Ed.,
              "Network Service Header (NSH)", RFC 8300,
              DOI 10.17487/RFC8300, January 2018,
              <https://www.rfc-editor.org/info/rfc8300>.

10.2.  Informative References



   [BGP-CONTROL]
              Farrel, A., Drake, J., Rosen, E., Uttaro, J., and L.
              Jalil, "BGP Control Plane for NSH SFC", Work in Progress,
              draft-ietf-bess-nsh-bgp-control-plane-04, July 2018.

   [PCEP-EXTENSIONS]
              Wu, Q., Dhody, D., Boucadair, M., Jacquenet, C., and J.
              Tantsura, "PCEP Extensions for Service Function Chaining
              (SFC)", Work in Progress,
              draft-wu-pce-traffic-steering-sfc-12, June 2017.

   [RADIUS]   Maglione, R., Trueba, G., and C. Pignataro, "RADIUS
              Attributes for NSH", Work in Progress,
              draft-maglione-sfc-nsh-radius-01, October 2016.

   [RFC3443]  Agarwal, P. and B. Akyol, "Time To Live (TTL) Processing
              in Multi-Protocol Label Switching (MPLS) Networks",
              RFC 3443, DOI 10.17487/RFC3443, January 2003,
              <https://www.rfc-editor.org/info/rfc3443>.







Dolson, et al.                Experimental                     [Page 22]

RFC 8459                          hSFC                    September 2018


   [USE-CASES]
              Kumar, S., Tufail, M., Majee, S., Captari, C., and S.
              Homma, "Service Function Chaining Use Cases In Data
              Centers", Work in Progress,
              draft-ietf-sfc-dc-use-cases-06, February 2017.














































Dolson, et al.                Experimental                     [Page 23]

RFC 8459                          hSFC                    September 2018


Appendix A.  Examples of Hierarchical Service Function Chaining



   The advantage of hSFC compared with normal or flat service function
   chaining is that it can reduce the management complexity
   significantly.  This section discusses examples that show those
   advantages.

A.1.  Reducing the Number of Service Function Paths



   In this case, hSFC is used to simplify service function chaining
   management by reducing the number of SFPs.

   As shown in Figure 6, there are two domains, each with different
   concerns: a Security Domain that selects SFs based on network
   conditions and an Optimization Domain that selects SFs based on
   traffic protocol.

   In this example, there are five security functions deployed in the
   Security Domain.  The Security Domain operator wants to enforce the
   five different security policies, and the Optimization Domain
   operator wants to apply different optimizations (either cache or
   video optimization) to each of these two types of traffic.  If we use
   flat SFC (normal branching), 10 SFPs are needed in each domain.  In
   contrast, if we use hSFC, only five SFPs in Security Domain and two
   SFPs in Optimization Domain will be required, as shown in Figure 7.

   In the flat model, the number of SFPs is the product of the number of
   SFs in all of the domains.  In the hSFC model, the number of SFPs is
   the sum of the number of SFs.  For example, adding a "bypass" path in
   the Optimization Domain would cause the flat model to require 15
   paths (five more) but cause the hSFC model to require one more path
   in the Optimization Domain.



















Dolson, et al.                Experimental                     [Page 24]

RFC 8459                          hSFC                    September 2018


              . . . . . . . . . . . .   . . . . . . . . . . . . .
              . Security Domain     .   .  Optimization Domain  .
              .                     .   .                       .
              .    +-1---[     ]----------------->[Cache  ]------->
              .    |     [ WAF ]    .   .                       .
              .    +-2-->[     ]----------------->[Video Opt.]---->
              .    |                .   .                       .
              .    +-3---[Anti ]----------------->[Cache  ]------->
              .    |     [Virus]    .   .                       .
              .    +-4-->[     ]----------------->[Video Opt.]---->
              .    |                .   .                       .
              .    +-5-->[     ]----------------->[Cache  ]------->
   [DPI]--->[CF]---|     [ IPS ]    .   .                       .
              .    +-6-->[     ]----------------->[Video Opt.]---->
              .    |                .   .                       .
              .    +-7-->[     ]----------------->[Cache  ]------->
              .    |     [ IDS ]    .   .                       .
              .    +-8-->[     ]----------------->[Video Opt.]---->
              .    |                .   .                       .
              .    +-9-->[Traffic]--------------->[Cache  ]------->
              .    |     [Monitor]  .   .                       .
              .    +-10->[       ]--------------->[Video Opt.]---->
              . . . . . . . . . . . .   . . . . . . . . . . . . .


   Legend:
      IDS: Intrusion Detection System
      IPS: Intrusion Prevention System
      WAF: Web Application Firewall
      DPI: Deep Packet Inspection

   The classifier must select paths that determine the combination of
   Security and Optimization concerns. 1:WAF+Cache, 2:WAF+VideoOpt,
   3:AntiVirus+Cache, 4:AntiVirus+VideoOpt, 5:IPS+Cache, 6:IPS+VideoOpt,
   7:IDS+Cache, 8:IDS+VideoOpt, 9:TrafficMonitor+Cache,
   10:TrafficMonitor+VideoOpt

                   Figure 6: Flat SFC (Normal Branching)













Dolson, et al.                Experimental                     [Page 25]

RFC 8459                          hSFC                    September 2018


        . . . . . . . . . . . . . . .    . . . . . . . . . . . . . . .
        .     Security Domain       .    .   Optimization Domain     .
        .                           .    .                           .
   [CF]---->[  [CF]    IBN      ]---------->[  [CF]   IBN         ]---->
        .    |                  ^   .    .  |                     ^  .
        .    +----->[ WAF ]-----+   .    .  +-->[ Cache ]---------+  .
        .    |                  |   .    .  |                     |  .
        .    +-->[Anti-Virus]---+   .    .  +-->[Video Opt]-------+  .
        .    |                  |   .    .                           .
        .    +----->[ IPS ]-----+   .    . . . . . . . . . . . . . . .
        .    |                  |   .
        .    +----->[ IDS ]-----+   .
        .    |                  |   .
        .    +-->[ Traffic ]----+   .
        .        [ Monitor ]        .
        . . . . . . . . . . . . . . .

              Figure 7: Simplified Path Management with hSFC

A.2.  Managing a Distributed DC Network



   Hierarchical service function chaining can be used to simplify inter-
   DC SFC management.  In the example of Figure 8, there is a central
   data center (Central DC) and multiple local data centers (Local DC#1,
   #2, #3) that are deployed in a geographically distributed manner.
   All of the data centers are under a single administrative domain.

   The central DC may have some service functions that the local DC
   needs, such that the local DC needs to chain traffic via the central
   DC.  This could be because:

   o  Some SFs are deployed as dedicated hardware appliances, and there
      is a desire to lower the cost (both CAPEX and OPEX) of deploying
      such SFs in all data centers.

   o  Some SFs are being trialed or introduced, or they otherwise handle
      a relatively small amount of traffic.  It may be cheaper to manage
      these SFs in a single central data center and steer packets to the
      central data center than to manage these SFs in all data centers.












Dolson, et al.                Experimental                     [Page 26]

RFC 8459                          hSFC                    September 2018


                   +-----------+
                   |Central DC |
                   +-----------+
                      ^  ^   ^
                      |  |   |
                  .---|--|---|----.
                 /   /   |   |      \
                /   /    |    \      \
     +-----+   /   /     |     \      \    +-----+
     |Local|  |   /      |      \     |    |Local|
     |DC#1 |--|--.       |       .----|----|DC#3 |
     +-----+  |          |            |    +-----+
               \         |            /
                \        |           /
                 \       |          /
                  '----------------'
                         |
                      +-----+
                      |Local|
                      |DC#2 |
                      +-----+


                Figure 8: Simplify Inter-DC SFC Management

   For large DC operators, one local DC may have tens of thousands of
   servers and hundreds of thousands of virtual machines.  SFC can be
   used to manage user traffic.  For example, SFC can be used to
   classify user traffic based on service type, DDoS state, etc.

   In such a large-scale DC, using flat SFC is very complex, requiring a
   super controller to configure all DCs.  For example, any changes to
   SFs or SFPs in the central DC (e.g., deploying a new SF) would
   require updates to all of the SFPs in the local DCs accordingly.
   Furthermore, requirements for symmetric paths add additional
   complexity when flat SFC is used in this scenario.

   Conversely, if using hierarchical SFC, each DC can be managed
   independently to significantly reduce management complexity.  SFPs
   between DCs can represent abstract notions without regard to details
   within DCs.  Independent controllers can be used for the top level
   (getting packets to pass the correct DCs) and local levels (getting
   packets to specific SF instances).








Dolson, et al.                Experimental                     [Page 27]

RFC 8459                          hSFC                    September 2018


Acknowledgements



   The concept of Hierarchical Service Path Domains was introduced in
   "Analysis on Forwarding Methods for Service Chaining" (August 2016)
   as a means of improving scalability of service chaining in large
   networks.

   The concept of nesting NSH headers within lower-level NSH was
   contributed by Ting Ao.  The concept originally appeared in
   "Hierarchical SFC for DC Interconnection" (April 2016) as a means of
   creating hierarchical SFC in a data center.

   We thank Dapeng Liu for contributing the DC examples in the Appendix.

   The Stateful/Metadata Hybrid section was contributed by Victor Wu.

   The authors would also like to thank the following individuals for
   providing valuable feedback:

      Ron Parker
      Christian Jacquenet
      Jie Cao
      Kyle Larose

   Thanks to Ines Robles, Sean Turner, Vijay Gurbani, Ben Campbell, and
   Benjamin Kaduk for their review.

























Dolson, et al.                Experimental                     [Page 28]

RFC 8459                          hSFC                    September 2018


Authors' Addresses



   David Dolson
   Sandvine
   Waterloo, ON
   Canada

   Email: ddolson@acm.org


   Shunsuke Homma
   NTT
   3-9-11, Midori-cho
   Musashino-shi, Tokyo  180-8585
   Japan

   Email: homma.shunsuke@lab.ntt.co.jp


   Diego R. Lopez
   Telefonica I+D
   Don Ramon de la Cruz, 82
   Madrid  28006
   Spain

   Phone: +34 913 129 041
   Email: diego.r.lopez@telefonica.com


   Mohamed Boucadair
   Orange
   Rennes  35000
   France

   Email: mohamed.boucadair@orange.com
















Dolson, et al.                Experimental                     [Page 29]