Internet DRAFT - draft-eastlake-sfc-parallel
draft-eastlake-sfc-parallel
SFC Working Group D. Eastlake
Internet-Draft Futurewei Technologies
Intended status: Standards Track 26 November 2023
Expires: 29 May 2024
Service Function Chaining (SFC) Parallelism and Diversions
draft-eastlake-sfc-parallel-07
Abstract
Service Function Chaining (SFC) is the processing of packets through
a sequence of Service Functions (SFs) within an SFC domain by the
addition of path information and metadata on entry to that domain,
the use and modification of that path information and metadata to
step the packet through a sequence of SFs, and the removal of that
path information and metadata on exit from that domain. The IETF has
standardized a method for SFC using the Network Service Header
specified in RFC 8300.
There are requirements for SFC to process packets through parallel
sequences of service functions, rejoining thereafter, and to easily
splice in additional service functions or splice service functions
out of a service chain. The IETF has received a liaison from
International Telecommunication Union (ITU) indicating their interest
in such requirements. This document provides use cases and specifies
extensions to SFC to support these requirements.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on 29 May 2024.
Eastlake Expires 29 May 2024 [Page 1]
Internet-Draft SFC Parallelism and Diversions November 2023
Copyright Notice
Copyright (c) 2023 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 Revised BSD License text as
described in Section 4.e of the Trust Legal Provisions and are
provided without warranty as described in the Revised BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1. Conventions Used in This Document . . . . . . . . . . . . 3
2. Service Function Chaining Background . . . . . . . . . . . . 4
2.1. The Network Service Header (NSH) . . . . . . . . . . . . 6
2.2. NSH Metadata and Variable Length Context Headers . . . . 7
3. Requirements for Parallelism and Diversions . . . . . . . . . 8
4. Diversion Points and Rendezvous Points . . . . . . . . . . . 11
4.1. Rendezvous Point Information (RePIn) . . . . . . . . . . 11
4.1.1. Packet Identifier . . . . . . . . . . . . . . . . . . 12
4.1.2. Packet Extent Modified . . . . . . . . . . . . . . . 13
4.1.3. Saved Metadata . . . . . . . . . . . . . . . . . . . 14
4.1.4. Saved TTL . . . . . . . . . . . . . . . . . . . . . . 15
4.2. Diversion Point (DP) Behavior . . . . . . . . . . . . . . 15
4.3. Rendezvous Point (RP) Behavior . . . . . . . . . . . . . 17
5. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 19
5.1. Variable Length Context Header Type . . . . . . . . . . . 19
5.2. RePIn VLCH Sub-Types . . . . . . . . . . . . . . . . . . 19
6. Security Consideration . . . . . . . . . . . . . . . . . . . 20
7. Normative References . . . . . . . . . . . . . . . . . . . . 20
8. Informative References . . . . . . . . . . . . . . . . . . . 20
Appendix A. Relation to Hierarchical SFC . . . . . . . . . . . . 21
Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 21
1. Introduction
Service Function Chaining (SFC) is the processing of packets through
a sequence of Service Functions (SFs) within an SFC domain by the
addition of path information and metadata on entry to that domain,
the use and modification of that path information and metadata to
step the packet through a sequence of SFs, and the removal of that
path information and metadata on exit from that domain. The IETF has
standardized a method for SFC using the Network Service Header
Eastlake Expires 29 May 2024 [Page 2]
Internet-Draft SFC Parallelism and Diversions November 2023
specified in [RFC8300].
There are requirements for SFC to process packets through parallel
sequences of service functions, rejoining thereafter, and to easily
splice in additional service functions or splice service functions
out of a service chain. The IETF has received a liaison from the ITU
[Liaison] indicating their interest in such requirements. This
document provides use cases and specifies extensions to SFC to
support these requirements.
1.1. Conventions Used in This Document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP
14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
Acronyms and terms:
downstream - The direction from ingress to egress.
diversion - A reclassification of an SFC packet into one or multiple
parallel packets with difference SPIs where this reclassified
packet or packets are, in the normal case, combined at a
downstream rendezvous point which restores the original SPI.
DP - Diversion Point - An SF implementing a diversion.
ITU - Iternational Telecommunications Union (www.itu.int).
MD - Metadata - Part of the NSH.
NSH - Network Service Header [RFC8300].
rendezvous - The process of taking one or more corresponding SFC
packets that have been diverted at an upstream DP, combining the
packets if there are more than one, and restoring the original
SPI.
RePIn - Rendezvous Point Information. Metadata included in an SFC
packet for use at an RP.
RP - Rendezvous Point - An SF implementing a rendezvous.
SF - Service Function [RFC7665].
SFC - Service Function Chaining [RFC7665].
Eastlake Expires 29 May 2024 [Page 3]
Internet-Draft SFC Parallelism and Diversions November 2023
SFF - Service Function Forwarder [RFC7665] - A type of node that
forwards packets based on the NSH.
SFP - Service Function Path.
SI - Service Index - Part of the NSH.
SPI - Service Path Identifier - Part of the NSH.
TLV - Type Length Value.
upstream - The direction from egress to ingress.
VLCH - Variable Length Context Header - A type of NSH header
metadata.
2. Service Function Chaining Background
Service Function Chaining (SFC) calls for the encapsulation of
traffic within a service function chaining domain using a Network
Service Header (NSH [RFC8300]) added by the "Classifier" (ingress
node) on entry to the domain and the NSH being removed on exit from
the domain at the downstream egress node as shown in Figure 1. The
NSH controls the path of a packet in an SFC domain and includes
additional information.
Eastlake Expires 29 May 2024 [Page 4]
Internet-Draft SFC Parallelism and Diversions November 2023
|
v
+----------+
. .|Classifier|. . . . . . . . . . . . . .
. +----------+ .
. | +----+ .
. | --+ SF | Service .
. | / +----+ Function .
. v --- Chaining .
. +-----+/ +----+ domain .
. | SFF |--------+ SF | .
. +-----+\ +----+ .
. | --- .
. | \ +----+ .
. | --+ SF | .
. v +----+ .
. +-----+ +----+ .
. | SFF |-----------------+ SF | .
. +-----+ +----+ .
. | +----+ .
. | --+ SF | .
. | / +----+ .
. v --- .
. +-----+/ +----+ .
. | SFF |--------+ SF | .
. +-----+\ +----+ .
. | --- .
. | \ +----+ .
. | --+ SF | .
. v +----+ .
. +------+ .
. . .| Exit |. . . . . . . . . . . . . . .
+------+
|
v
Figure 1: Example SFC Path Forwarding Nodes
Traffic passes through a sequence of Service Function Forwarders
(SFFs) each of which sends the traffic to one or, sequentially, more
than one Service Functions (SFs). Each SF performs some operation on
the traffic, for example firewall or Network Address Translation
(NAT) or load balancer, and then returns it to the SFF from which it
was received. There may be multiple instances of SFs performing the
same function attached to the same or different SFFs.
Eastlake Expires 29 May 2024 [Page 5]
Internet-Draft SFC Parallelism and Diversions November 2023
Logically, during the transit of an SFF, the outer transport header
that got the packet to the SFF is stripped (see Figure 2), the SFF
decides on the next forwarding step either (1) adding a new transport
header or (2) in case of error discarding or logging the packet and
not forwarding it or (3) if the SFF is the exit/egress, removing the
NSH header and then adding a new transport header. The transport
used may be different in different regions of the SFC domain. For
example, a version of the Internet Protocol (IP) could be used in
some parts and Multi-Protocol Label Switching (MPLS) used in other
parts of the SFC domain.
+-----------------------------------+
| Outer Transport Header |
+-----------------------------------+
| Network Service Header (NSH) |
| +------------------------------+ |
| | Base Header | |
| +------------------------------+ |
| | Service Path Header | |
| +------------------------------+ |
| | Metadata (Context Header(s)) | |
| +------------------------------+ |
+-----------------------------------+
| Original Packet / Frame / Payload |
+-----------------------------------+
Figure 2: Data Encapsulation with the NSH
An SF can receive one or more SFC packets from an SFF and return to
it a larger or smaller number of SFC packets; that is to say, SFC
packets can be discarded or created by an SF.
2.1. The Network Service Header (NSH)
The NSH header is used to encapsulate and control the subsequent path
of traffic. It consists of three parts, the initial 32-bit Base
Header, the 32-bit Service Path Header, and any Context Headers
holding metadata, as shown in more detail in Figure 3 and specified
in [RFC8300].
The Base Header includes a Length field whose value is the overall
NSH length. Because the Base Header and Service Path Header are
fixed length, the length of the Metadata can be computed from this
Length field. The Base Header also includes a field indicating the
type of metadata in the NSH.
Eastlake Expires 29 May 2024 [Page 6]
Internet-Draft SFC Parallelism and Diversions November 2023
The Service Path Header consists of a Service Path Identifier (SPI)
and a Service Index (SI). The SPI identifies the logical path the
packet should follow while the SI indicates which step along that
path the packet is at.
An SF anywhere along a Service Function Path can re-classify an SFC
packet by replacing the Service Path Identifier (SPI) and Service
Index (SI) in the NSH. SFFs can also insert, delete, or change
metadata (Context Header(s)) in the NSH.
0 1 2 3
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 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Optional Context Headers / Metadata ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
Figure 3: Network Service Header Details from [RFC8300]
2.2. NSH Metadata and Variable Length Context Headers
If the MD Type field in the NSH Base Header has the value 1, there is
a single fixed length 128-bit Context Header whose format is not
further defined by the IETF. In that case, the NSH Length field has
the value 6.
If the MD Type field has the value 2, there are zero or more
Variable-Length Context Headers (VLCHs) as shown in Figure 4 at the
end of the NSH. The absence of any Context Headers is indicated by
using MD Type 2 and an NSH Length of 2. MD Type 0 is reserved and MD
Types 3 through 15 are unassigned.
0 1 2 3
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Metadata Class | Type |U| Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Variable Length Metadata ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
Figure 4: Variable Length Context Header
The minimum size for a VLCH is 32 bits consisting of the Metadata
Class, Type, one unused bit, and Length as shown in Figure 4.
Eastlake Expires 29 May 2024 [Page 7]
Internet-Draft SFC Parallelism and Diversions November 2023
The Metadata Class field is 16 bits, and its value specifies the
organization under which the particular of VLCHs specified. Metadata
Class zero is the IETF base class. The 8-bit Type field's value,
along with the Metadata Class value, indicates the meaning of the
Context Header and its Variable-Length Metadata. The size of the
Length field is 7 bits and its value gives, as an unsigned integer,
the length in octets of the Variable-Length Metadata that follows the
initial fixed length portion of the VLCH. A VLCH with no Variable-
Length Metadata is indicated by a Length field whose value is zero.
VLCHs are padded so that they always start and end at a multiple of 4
bytes from the beginning of the NSH.
3. Requirements for Parallelism and Diversions
There are requirements to split a Service Function Chain (SFC) into
two or more parallel Service Function Paths (SFPs) that later rejoin
as shown in Figure 5.
+------+ +-----+
----->|SFF 2a|<---->| SF 2|
/ +------+\ +-----+
/ \____
/ \
+-----+/ +------+ ->+-----+
---->|SFF 1|--------->|SFF 2b|-------->|SFF 3|----->
+-----+\ +------+ ->+-----+
/|\ \ /|\ / /|\
| \ | / |
\|/ \ \|/ | \|/
+-----+ \ +-----+ | +-----+
| SF 1| \ | SF 3| / | SF 5|
| DP | \ +-----+ / | RP |
+-----+ \ / +-----+
\ +------+/ +-----+
->|SFF 2c|<---->| SF 4|
+------+ +-----+
Figure 5: Parallel Service Function Paths
For example, there may be two or more Service Functions (SFs) that
can be performed in parallel with the goal, for time critical traffic
such as some financial or gaming traffic, of delaying the stream of
packets only by the amount of time taken by the slowest of those SFs;
if the packets went through the SFs sequentially, the delay would be
the sum of the times taken by each of the SFs. An example of such
potential parallel processing might be that the SFs operate of
different parts of the packet such as one SF operating on packet
addressing while another operates on the information payload.
Eastlake Expires 29 May 2024 [Page 8]
Internet-Draft SFC Parallelism and Diversions November 2023
Another example might be that one SF creates a signature or integrity
code over parts of the packet to be inserted into the packet payload
while another SF encrypts parts of the packet (or alternatively, they
verify and decrypt in parallel). As indicated by their [Liaison],
the ITU is interested in such use cases.
Another example of desirable parallelism would be improved
reliability or accuracy if the SFs executed in parallel were
unreliable or were different implementations of doing the same
processing. For example, some quantum computers are currently
unreliable so it would be desirable to perform some quantum process
several times and compare the results to pick the most common value,
or a vote could be taken between the results of different
implementations of some process.
In Figure 5 it could be that any of the parallel paths could have
more or less than one SFF, although exactly one is shown in the
example for simplicity and any of the SFFs in any of the parallel
SFPs could process a packet through more than one SF, although they
are shown using only one SF in Figure 5 for simplicity. (Note that
while SFFs implement an SFP, SFPs logically consists of the sequence
of SFs. Thus, for example, an SFP could divert into multiple
parallel SFPs that rejoin at an RP all implemented through SFs off of
one and the same SFF.) It could also be that one or more of the
parallel paths would themselves further split into parallel paths and
so on.
There are cases where it is desirable to divert an SFP so as to
splice one or more added SFs into that SFP or to divert it so as to
slice out one or more sequential SFs that were downstream in that
SFP, as shown in Figure 6 and Figure 7. Although SF 3 in each of
those two cases could re-classify the packet with a new SPI and SI
that include the remainder of the new diverted path, this would
require that a new SFP with this new SPI already be configured in all
the SFFs for the remainder of the Service Function Path after a
diversion. In the case of Figure 7, SF 3 could possibly just adjust
the Service Index, but this would require relaxing any checking at SF
5 of the SPI/SI or source address of packets on the main SFP or may
otherwise be undesirable.
Eastlake Expires 29 May 2024 [Page 9]
Internet-Draft SFC Parallelism and Diversions November 2023
+-----+ +-----+
->|SFF 3| <---->| SF 4|
/ +-----+\ +-----+
/ \
+-----+ +-----+ ->+-----+ +-----+
--->|SFF 1|----->|SFF 2| - - - - - - >|SFF 4|----->|SFF 5|--->
+-----+ +-----+ +-----+ +-----+
/|\ /|\ /|\ /|\
| | | |
\|/ \|/ \|/ \|/
+-----+ +-----+ +-----+ +-----+
| SF 1| | SF 3| | SF 5| | SF 6|
+-----+ | DP | | RP | +-----+
+-----+ +-----+
Figure 6: Splicing in One or More SFFs
------------\
/ \
/ \
+-----+ +-----+ +-----+ \->+-----+
--->|SFF 1|----->|SFF 2|- - ->|SFF 4|- - ->|SFF 5|--->
+-----+ +-----+ +-----+ +-----+
/|\ /|\ /|\ /|\
| | | |
\|/ \|/ \|/ \|/
+-----+ +-----+ +-----+ +-----+
| SF 1| | SF 3| | SF 5| | SF 5|
+-----+ | DP | +-----+ | RP |
+-----+ +-----+
Figure 7: Splicing out One or More SFFs
Combinations of the cases shown in Figure 5, Figure 6, and Figure 7
may be needed where a diversion such as in Figure 6 or Figure 7 occur
within a parallel path as in Figure 5 or parallel paths as in
Figure 5 occur within a diversion as in Figure 6. Generalizing
Figure 6 and Figure 7, a diversion might splice in a path with some
number of SFs that cuts out a portion of the original SFP that had
some number of SFs.
Although DPs and RPs are logically Service Functions (SFs) and shown
as separate boxes in the above figures, like any other SF they can be
implemented as co-located with an SFF.
Eastlake Expires 29 May 2024 [Page 10]
Internet-Draft SFC Parallelism and Diversions November 2023
4. Diversion Points and Rendezvous Points
SF 1 in Figure 5 and SF 3 in Figure 6 and Figure 7 are referred to as
Diversion Points (DPs) because they are nodes at which an SFP is
diverted to one or more SFPs with new SPIs that are intended to
rejoint/return to the original SPI at a downstream Rendezvous Point.
SF 5 in Figure 5, Figure 6, and Figure 7 is referred to as a
Rendezvous Point (RP) because it is the node at which one or more
SFPs from an upstream DP rejoin an original SFP and an original SPI
is restored. The coresponding packets so received at an RP are
merged or coordinated.
In general, an RP needs to be configured to expect SFC packets to
arrive at that RP on diverted SFPs. An RP may need additional
information in the SFC packets, as discussed in Section 4.1, to be
included in their NSH. Divergence point behavior is discussed in
Section 4.2 and RP behavior is discussed in Section 4.3.
4.1. Rendezvous Point Information (RePIn)
To recombine packets from divergent SFP(s) at a Rendezvous Point
(RP), or rejoin a diverted SFP to the original SFP, additional
information may be needed in the packets. This is accomplished
through inclusion of the RP Information (RePIn) Variable Length
Context Header (VLCH), as shown in Figure 8, in the packet's NSH.
0 1 2 3
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Metadata Class = 0x0000 | Type=TBD |U| Length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Saved Service Path Identifier | Saved SI |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Variable Length Sub-TLVs ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
Figure 8: RePIn VLCH
The Saved Service Path Identifier and Saved SI are the SPI and SI in
the NSH of the SFC packet being diverted after entry to the DP SF and
the SI has been decremented.
The Length field is the total length of the Variable Length Sub-TLVs
in octets plus 4 for the length of the Saved SPI and SI.
Eastlake Expires 29 May 2024 [Page 11]
Internet-Draft SFC Parallelism and Diversions November 2023
The Variable Length Sub-TLVs consist of zero or more RePIn VLCH Sub-
TLVs. The format of a RePIn VLCH Sub-TLV is as shown in Figure 9
except for Sub-Type 1 as discussed in Section 4.1.1; however, all
RePIn VLCH Sub-TLVs are padded at the end up to an even multiple of 4
octets.
0 1 2 3
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Sub-Type |X|Sub-Length | Variable Length Metadata ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
Figure 9: General RePIn VLCH Sub-TLV
The Sub-Type field is an 8-bit unsigned integer that is always
present and indicates the format of the Variable Length Metadata in
the Sub-TLV. The X bit may be assigned a meaning for particular Sub-
Types; if no such meaning is assigned for a particular Sub-Type, the
X bit MUST be sent as zero and ignored on receipt. Sub-Length is an
unsigned integer giving the length of the variable length metadata in
octets.
Unless the specification for a RePIn VLCH Sub-TLV Sub-Type specifies
that there may be multiple occurrences of that Sub-TLV, it may only
be included once. If there are multiple instances, the first
occurrence is used with any subsequent occurrences being ignored.
4.1.1. Packet Identifier
When an SFC packet is replicated and diverted to more than one
parallel path to be merged back together at a Rendezvous Point (RP),
a method of matching packets is needed such as labeling each copy
that originated with the same packet before the split using a packet
identifier such as a packet counter, fine grained time stamp, or hash
code. Such an identifier might already exist in the packet, for
example a TCP sequence number. If such an existing identifier cannot
be trusted or there is none, the Packet Identifier sub-TLV as shown
below is included as a VLCH. Use of the Packet Counter sub-TLV is
RECOMMENDED.
0 1 2 3
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Sub-Type=1 | Packet Identifier |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 10: Packet Identifier Special Sub-TLV
Eastlake Expires 29 May 2024 [Page 12]
Internet-Draft SFC Parallelism and Diversions November 2023
Because this is expected to be a common RePIn VLCH Sub-TLV, in order
to save octets, for this Sub-Type only, the "Sub-TLV" X and Sub-
Length fields are subsumed into the Packet Identifier field.
4.1.2. Packet Extent Modified
If two or more SFPs used in parallel have modified parts of a packet,
the RP may need additional information to be able to recombine the
different copies of the packet it will be receiving. As an example
of the complexities involved, an SF could change the length of part
of a packet in a way dependent on the content of that part such as by
applying a data compression or de-compression algorithm to part of
the packet or by conditionally inserting or removing a VLAN tag
depending on addressing information.
In simple cases such as parallel SFPs that modify fixed size disjoint
parts of a packet without changing the size of those parts, it may be
possible for an RP to be configured to recombine the results without
added information. But in more complex or variable length cases,
parallel SFPs need to add information as to what part of the original
packet they modified and how this may have changed the length of that
part. Also, with such additional information, in some cases only one
of the parallel SFPs would need to forward all of the original packet
with modifications to the RP; one or more other parallel SFPs could
just forward their modified part and the RP would be able to
recombine the results thus saving communications link capacity that
would be used if they all sent full packets.
0 1 2 3
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Sub-Type=2 |X|Sub-Length=6 | Offset |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Priority | Original Size | Modified Size |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 11: Packet Extent Modified Sub-TLV
If the X bit is zero, the entire modified packet is present in the
SFC packet. If the X bit is a one, only the modified portion appears
in the packet which requires any SFs between the SF that modified the
packet and the RP to be capable of handling such an abbreviated
packet.
Eastlake Expires 29 May 2024 [Page 13]
Internet-Draft SFC Parallelism and Diversions November 2023
Offset is the number of bits between the end of the NSH or the last
NSH if there are multiple stacked NSHs and the portion of the packet
being modified. Original Size and Modified Size are the size in bits
of the portion being modified before and after that modification.
Any of the Offset, Original Size, and Modified Size fields may have
the value zero.
If parallel SFPs have modified the same or overlapping parts of a
packet, the RP may need some way to resolve this conflict which could
include a relative priority for changes made by different SFs
configured at the RP and/or indicated in the RP Information (RePIn)
or from other sources. The Priority field may be used for this
purpose; it contains an unsigned integer where a larger magnitude
value indicates a higher priority that would prevail over a lower
priority. If not used, the Priority field MUST be sent as zero and
ignored on receipt.
If a path has modified more than one portion of a packet, multiple
instance of the Packet Extent Modified Sub-TLV can be included in the
RePIn VLCH. If any Packet Extent Modified Sub-TLV Sub-Lenth is any
value other than 6, the meta-data is corrupt, the packet is silently
discarded, and an error SHOULD be logged.
4.1.3. Saved Metadata
A DP may need to save Metadata so it will not be seen inside a
diversion and will be restored at the RP. This Sub-TLV is used for
that purpose. See Section 4.2 and 4.3 for further details on the use
of this sub-TLV.
0 1 2 3
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Sub-Type=3 |X| Sub-Length | MBZ |MetaTyp|
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Variable Length Saved Metadata ~
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
Figure 12: Saved Metadata Sub-TLV
The X and MBZ bits MUST be sent as zero and ignored on receipt.
MetaTyp is the MD Type of Metadata saved. The presence of the MBZ
field causes the saved metadata to be aligned on a multiple of 4
octets.
Eastlake Expires 29 May 2024 [Page 14]
Internet-Draft SFC Parallelism and Diversions November 2023
4.1.4. Saved TTL
The TTL limits the number of SFs that can be traversed between
ingress and egress. The packet is discarded if the TTL is exhausted.
This is a safety measure to defend against infinite or very large
loops due to malfunctions, configuration error, or other reasons.
Thus, the RECOMMENDED mode of operation is to use a TTL value that is
decremented continuously from original SFC domain ingress to final
SFC egress including throughout any diversions. If the TTL is reset
on entry to a diversion, then the Saved TTL Sub-TLV MUST be used so
that the previous TTL can be restored at the diversion's RP.
Note that resetting the TTL on entry to a diversion opens the
possibility for loop where a diversion diverts to itself or there are
two diversions X and Y where X diverts to Y and Y diverts to X or
more complex scenarios all of which are made safe by using a
continuous TTL and unsafe by resetting the TTL on diversion entry.
Such loops will result in a growing amount of metadata which might
safely lead to packet discard or unsafely cause repeated
fragmentation.
If, despite the above warning, it is desired to reset the TTL at the
DP and restore it at the RP, the Saved TTL Sub-TLV as shown below is
used.
0 1 2 3
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Sub-Type=4 |X| Sub-Length=2| MBZ | Saved TTL |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 13: Saved TTL Sub-TLV
The X and MBZ bits MUST be sent as zero and ignored on receipt. The
Saved TTL is the value of the TTL field copied from the NSH after its
initial decrement on entering the DP SF. If the Saved TTL Sub-TLV
Sub-Lenth is any value other than 2, the meta-data is corrupt, the
packet is silently discarded, and an error SHOULD be logged.
4.2. Diversion Point (DP) Behavior
If it is desired to simply skip some SFs in an SFP, a diversion may
not be necessary. The SI can simply be decreased to that for the
next SF desired if the SFF to which the SF/DP that reduces the SI
returns the packet can handle that reduced SI value and forward the
packet to the appropriate SFF. Otherwise, the procedure below in
this section is used and this procedure MAY be used even in cases
where simple reduction of the SI would work.
Eastlake Expires 29 May 2024 [Page 15]
Internet-Draft SFC Parallelism and Diversions November 2023
If the RP can recognize diverted SFC packets and modify/merge them
appropriately to restore them to the original SFP with appropriate
Metadata, then inserting a RePIn VLCH might not be needed. In other
cases take the steps below. This is a logical procedure and any
procedure can be used that results in the same diverted SFC
packet(s).
1. Construct a RePIn VLCH containing the SPI and SI from the NSH and
then change those fields in the NSH to the diversion SPI and SI.
If diversion is to multiple parallel SFPs, make a copy of the SFC
packet for each diversion, then construct a VLCH and modify the
NSH as in the previous sentence for each one of the parallel
paths. Then perform the following steps to one or more modified
copies and its corresponding RePIn VLCH.
2. If it will be necessary for the RP to merge the modified copies
of the original SFC packet sent over parallel paths or if the RP
needs to restore a particular ordering to packets, then add a
Packet Identification Sub-TLV to the RePIn VLCH unless sufficient
trusted information is available in the packet payload for the RP
to do so without a Packet Identification Sub-TLV.
3. Take any Metadata from the original NSH that should be hidden
during the diversion and restored at the RP and add it to the
RePIn VLCH as a Saved Metadata Sub-TLV. If the NSH already has
any RePIn VLCHs, they need not be saved as they will be masked by
the new RePIn VLCH that will be inserted before them (this
indicates that a diversion from a diversion is being created).
To save space, any Metadata that has been saved in the RePIn VLCH
and is not needed in the diversion SFP SHOULD be removed from the
NSH if MD Type 2 and MUST be removed from the NSH if MD Type 1.
(In the Type 1 case, this converts the NSH to MD Type 2 with no
Metadata.)
4. If it is desired to use a new value for the NSH TTL in the
diversion, with the old value restored at the RP, add a Saved TTL
Sub-TLV to the RePIn VLCH and set the TTL in the NSH to a
configured value which may be dependent on the diversion being
entered. This is NOT RECOMMENDED as discussed in Section 4.1.4.
5. The RePIn VLCH constructed as above is inserted into an NSH as in
the subpoints below:
5.a. If, after the above step 3, the initial NSH in the SFC
packet is MD Type 2, insert the RePIn VLCH constructed above
before any existing RePIn VLCH that may be in the NSH.
Eastlake Expires 29 May 2024 [Page 16]
Internet-Draft SFC Parallelism and Diversions November 2023
5.b. If, after the above step 3, the initial NSH in the SFC
packet is MD Type 1, this implies that there is Type 1 metadata
that may be needed by one or more SFs in the diversion. If the
SFs in the diversion can handle stacked NSHs, insert an MD Type 2
NSH copied from the initial NSH except for metadata, after the
initial MD Type 1 NSH to hold the RePIn VLCH. Handling stacks
NSHs means the SF (or its proxy) can parse through them to find
the needed metadata and the payload to operate on and, if the SF
generates packets, can create them with appropriate stacked NSHs.
If the SFs in the diversion cannot handle stacked NSHes, the
creation of the diversion is beyond the scope of this document.
6. Perform such other functions or modifications to the metadata or
other parts of the SFC packet as are appropriate based on the
saved or new SPI or other factors.
7. Transmit the modified packets.
The addition of Metadata and possible additional NSH header (see step
5.b above) may lead to fragmentation or decreased payload Maximum
Transmission Unit (MTU) in some networks.
4.3. Rendezvous Point (RP) Behavior
A RP will have been configured to know the SFC packet SPI and SI
values in diverted packets for which it is to perform the RP service.
The SI is decremented when an SFC packet is received by an SF; for an
RP this might decrement the SI to zero. The RP performs the steps
below. If the RP can restore diverted SFC packets to their former
SFP and, to the extent necessary, match and merge diverted packets
received over parallel paths and correctly order the resulting SFC
packets, without the presence of a RePIn VLCH, it does so and the
remainder of this section is inapplicable. If not, the following
logical procedure or any procedure resulting in the same SFC packet
is used.
1. Steps 2 and 3 below are performed on each diverted packet
received by the RP. If the RP is merging parallel diversions,
step 4 is then performed on the set of matching packets. In this
case and any case where the RP should restore packet order, the
RP SHOULD be prepared to buffer packets until they can be
processed and forwarded. Overflow of such a buffer will result
in lost packets and SHOULD be logged as an error. How long to
wait for missing diverted packets and what action to take if it
is decided they have been lost are application and implementation
dependent. Finally, Step 5 is performed.
Eastlake Expires 29 May 2024 [Page 17]
Internet-Draft SFC Parallelism and Diversions November 2023
2. Find and remove the first RePIn VLCH in the diverted packet.
This is referred to below as the removed VLCH. It might be in a
second stacked NSH if the initial NSH has MD Type 1.
3. Restore the packet from the diversion through the sub-steps
listed below.
3.a. If there is a Saved TTL in the removed VLCH, restore the
old TTL into the initial NSH.
3.b. Restore the saved SPI and SI from the removed RePIn VLCH
into the initial NSH.
3.c. Restore metadata as follows:
3.c.1. Restore any MD Type 2 Saved Metadata from the removed
VLCH into the NSH from which that VLCH was removed.
3.c.2. If there is MD Type 1 Saved Metadata in the removed VLCH
and there is an initial MD Type 1 NSH in the packet, replace the
MD Type 1 metadata with the saved MD Type 1 metadata.
3.c.3. If there is MD Type 1 Saved Metadata in the removed VLCH
and there is an initial MD Type 2 NSH in the packet, insert a new
initial NSH into the packet which is a copy of that MD Type 2 NSH
except that it is MD Type 1 with the saved MD Type 1 metadata.
3.d. If, at this point, the packet starts with an MD Type 1 NSH
followed by an MD Type 2 NSH with no metadata, remove that 2nd
NSH.
4. Match up SFC packets arriving at the RP through parallel paths
using the Packet Identification Sub-TLV in the removed RePIn VLCH
or using some other technique. For each matching set, perform
the sub-steps below. Arbitrarily select one of the matching
diverted packets to modify into the merged packet unless
configured to use some particular diverted packet such as the one
received over a particular diversion. This is referred to below
as the merged packet even before the merger is complete.
4.a For error checking, any saved SPI and SI in the matching
packets SHOULD be compared and an error logged if they are not
identical.
4.b For safety, it is RECOMMENDED that the minimum of the NSH
TTLs from the parallel SFC packets be copied into the merged
packet.
Eastlake Expires 29 May 2024 [Page 18]
Internet-Draft SFC Parallelism and Diversions November 2023
4.c Depending on the application and implementation, the
remaining metadata in the merged packet may be used or updated
based on the remaining Metadata in the other packets being
merged. How to do this is beyond the scope of this document.
4.d The payloads of the other packets being merged, that is the
portion after any NSHs, are used to update the payload in the
merged packet. This may be based on RP configuration for the
application or Packet Extent Modified Sub-TLVs in the removed
RePIn VLCHs or a combination of these.
5. Perform such other functions or modifications to the metadata or
other parts of the SFC packet as are appropriate based on the
saved or new SPI or other factors.
6. Trasnmit the merged packet.
5. IANA Considerations
The following subsections provide IANA assignment considerations.
5.1. Variable Length Context Header Type
IANA is requested to assign a variable length context header type
from the "NSH IETF-Assigned Optional Variable-Length Metadata Types"
registry as follows:
+=======+======================================+=================+
| Value | Description | Reference |
+=======+======================================+=================+
| TDB | Rendezvous Point Information (RePIn) | [this document] |
+-------+--------------------------------------+-----------------+
Table 1
5.2. RePIn VLCH Sub-Types
IANA is requested to create a sub-registry under the "NSH IETF-
Assigned Optional Variable-Length Metadata Types" registry as
follows:
Name: Sub-TLVs under the Type TBD Variable Length Context Header
Registration Procedure: Expert Review
Reference: [this document]
Eastlake Expires 29 May 2024 [Page 19]
Internet-Draft SFC Parallelism and Diversions November 2023
+==========+========================+=================+
| Sub-Type | Description | Reference |
+==========+========================+=================+
| 0 | reserved | [this document] |
+----------+------------------------+-----------------+
| 1 | Packet Identifier | [this document] |
+----------+------------------------+-----------------+
| 2 | Packet Extent Modified | [this document] |
+----------+------------------------+-----------------+
| 3 | Saved Metadata | [this document] |
+----------+------------------------+-----------------+
| 4 | Saved TTL | [this document] |
+----------+------------------------+-----------------+
| 5-254 | unassigned | [this document] |
+----------+------------------------+-----------------+
| 255 | reserved | [this document] |
+----------+------------------------+-----------------+
Table 2
6. Security Consideration
For general SFC and NSH security considerations, see [RFC8300].
More TBD...
7. 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,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/info/rfc8174>.
[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>.
8. Informative References
[Liaison] "LS on recent service function chaining related
developments in Q4/SG11: two new draft Supplements",
ITU-T SG11-LS-179, March 2021,
<https://datatracker.ietf.org/liaison/1736/>.
Eastlake Expires 29 May 2024 [Page 20]
Internet-Draft SFC Parallelism and Diversions November 2023
[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>.
[RFC8459] Dolson, D., Homma, S., Lopez, D., and M. Boucadair,
"Hierarchical Service Function Chaining (hSFC)", RFC 8459,
DOI 10.17487/RFC8459, September 2018,
<https://www.rfc-editor.org/info/rfc8459>.
Appendix A. Relation to Hierarchical SFC
Experimental [RFC8459] describes "Hierarchical SFC" in which SFs in a
higher level SFP can be entire lower level SFPs with a different SPI
and where the higher level SPI is restored at the end of the lower
level SFP. This is similar to a diversion in this document. The
Internal Boundary Nodes (IBNs) in [RFC8459] that transition an SFC
packet between the higher and lower levels are similar to DPs/RPs in
the terminology of this document.
Experimental [RFC8459] discusses a wide variety of mechanisms to
implement Hierarchical SFC while this document looks toward
specifying a more specific set of mechanisms as a Proposed Standard
to support parallelism and other types of diversions.
Author's Address
Donald E. Eastlake 3rd
Futurewei Technologies
2386 Panoramic Circle
Apopka, Florida 32703
United States of America
Phone: +1-508-333-2270
Email: d3e3e3@gmail.com, donald.eastlake@futurewei.com
Eastlake Expires 29 May 2024 [Page 21]
