This is a purely informative rendering of an RFC that includes verified errata. This rendering may not be used as a reference.
The following 'Verified' errata have been incorporated in this document:
EID 3499, EID 3640, EID 4098
Internet Engineering Task Force (IETF) J. Macker, Ed.
Request for Comments: 6621 NRL
Category: Experimental May 2012
ISSN: 2070-1721
Simplified Multicast Forwarding
Abstract
This document describes a Simplified Multicast Forwarding (SMF)
mechanism that provides basic Internet Protocol (IP) multicast
forwarding suitable for limited wireless mesh and mobile ad hoc
network (MANET) use. It is mainly applicable in situations where
efficient flooding represents an acceptable engineering design trade-
off. It defines techniques for multicast duplicate packet detection
(DPD), to be applied in the forwarding process, for both IPv4 and
IPv6 protocol use. This document also specifies optional mechanisms
for using reduced relay sets to achieve more efficient multicast data
distribution within a mesh topology as compared to Classic Flooding.
Interactions with other protocols, such as use of information
provided by concurrently running unicast routing protocols or
interaction with other multicast protocols, as well as multiple
deployment approaches are also described. Distributed algorithms for
selecting reduced relay sets and related discussion are provided in
the appendices. Basic issues relating to the operation of multicast
MANET border routers are discussed, but ongoing work remains in this
area and is beyond the scope of this document.
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 a candidate for any level of
Internet Standard; see Section 2 of RFC 5741.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
http://www.rfc-editor.org/info/rfc6621.
Copyright Notice
Copyright (c) 2012 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
This document may contain material from IETF Documents or IETF
Contributions published or made publicly available before November
10, 2008. The person(s) controlling the copyright in some of this
material may not have granted the IETF Trust the right to allow
modifications of such material outside the IETF Standards Process.
Without obtaining an adequate license from the person(s) controlling
the copyright in such materials, this document may not be modified
outside the IETF Standards Process, and derivative works of it may
not be created outside the IETF Standards Process, except to format
it for publication as an RFC or to translate it into languages other
than English.
Table of Contents
1. Introduction and Scope ..........................................4
2. Terminology .....................................................4
3. Applicability Statement .........................................5
4. Overview and Functioning ........................................6
5. SMF Packet Processing and Forwarding ............................8
6. SMF Duplicate Packet Detection .................................10
6.1. IPv6 Duplicate Packet Detection ...........................11
6.1.1. IPv6 SMF_DPD Option Header .........................12
6.1.2. IPv6 Identification-Based DPD ......................14
6.1.3. IPv6 Hash-Based DPD ................................16
6.2. IPv4 Duplicate Packet Detection ...........................17
6.2.1. IPv4 Identification-Based DPD ......................18
6.2.2. IPv4 Hash-Based DPD ................................19
7. Relay Set Selection ............................................20
7.1. Non-Reduced Relay Set Forwarding ..........................20
7.2. Reduced Relay Set Forwarding ..............................20
8. SMF Neighborhood Discovery Requirements ........................23
8.1. SMF Relay Algorithm TLV Types .............................24
8.1.1. SMF Message TLV Type ...............................24
8.1.2. SMF Address Block TLV Type .........................25
9. SMF Border Gateway Considerations ..............................26
9.1. Forwarded Multicast Groups ................................27
9.2. Multicast Group Scoping ...................................28
9.3. Interface with Exterior Multicast Routing Protocols .......29
9.4. Multiple Border Routers ...................................29
10. Security Considerations .......................................31
11. IANA Considerations ...........................................32
11.1. IPv6 SMF_DPD Header Extension Option Type ................33
11.2. TaggerId Types (TidTy) ...................................33
11.3. Well-Known Multicast Address .............................34
11.4. SMF TLVs .................................................34
11.4.1. Expert Review for Created Type Extension
Registries ........................................34
11.4.2. SMF Message TLV Type (SMF_TYPE) ...................34
11.4.3. SMF Address Block TLV Type (SMF_NBR_TYPE) .........35
11.4.4. SMF Relay Algorithm ID Registry ...................35
12. Acknowledgments ...............................................36
13. References ....................................................37
13.1. Normative References .....................................37
13.2. Informative References ...................................38
Appendix A. Essential Connecting Dominating Set (E-CDS)
Algorithm ............................................40
A.1. E-CDS Relay Set Selection Overview ........................40
A.2. E-CDS Forwarding Rules ....................................41
A.3. E-CDS Neighborhood Discovery Requirements .................41
A.4. E-CDS Selection Algorithm .................................44
Appendix B. Source-Based Multipoint Relay (S-MPR) Algorithm ......46
B.1. S-MPR Relay Set Selection Overview ........................46
B.2. S-MPR Forwarding Rules ....................................47
B.3. S-MPR Neighborhood Discovery Requirements .................48
B.4. S-MPR Selection Algorithm .................................50
Appendix C. Multipoint Relay Connected Dominating Set
(MPR-CDS) Algorithm ..................................52
C.1. MPR-CDS Relay Set Selection Overview ......................52
C.2. MPR-CDS Forwarding Rules ..................................53
C.3. MPR-CDS Neighborhood Discovery Requirements ...............53
C.4. MPR-CDS Selection Algorithm ...............................54
1. Introduction and Scope
Unicast routing protocols, designed for MANET and wireless mesh use,
often apply distributed algorithms to flood routing control plane
messages within a MANET routing domain. For example, algorithms
specified within [RFC3626] and [RFC3684] provide distributed methods
of dynamically electing reduced relay sets that attempt to
efficiently flood routing control messages while maintaining a
connected set under dynamic topological conditions.
Simplified Multicast Forwarding (SMF) extends the efficient flooding
concept to the data forwarding plane, providing an appropriate
multicast forwarding capability for use cases where localized,
efficient flooding is considered an effective design approach. The
baseline design is intended to provide a basic, best-effort multicast
forwarding capability that is constrained to operate within a single
MANET routing domain.
An SMF routing domain is an instance of an SMF routing protocol with
common policies, under a single network administration authority.
The main design goals of this document are to:
o adapt efficient relay sets in MANET environments [RFC2501], and
o define the needed IPv4 and IPv6 multicast duplicate packet
detection (DPD) mechanisms to support multi-hop packet forwarding.
2. Terminology
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
[RFC2119].
The terms introduced in [RFC5444], including "packet", "message",
"TLV Block", "TLV", and "address", are to be interpreted as described
therein.
The following abbreviations are used throughout this document:
+--------------+----------------------------------------------------+
| Abbreviation | Definition |
+--------------+----------------------------------------------------+
| MANET | Mobile Ad Hoc Network |
| SMF | Simplified Multicast Forwarding |
| CF | Classic Flooding |
| CDS | Connected Dominating Set |
| MPR | Multipoint Relay |
| S-MPR | Source-based MPR |
| MPR-CDS | MPR-based CDS |
| E-CDS | Essential CDS |
| NHDP | Neighborhood Discovery Protocol |
| DPD | Duplicate Packet Detection |
| I-DPD | Identification-based DPD |
| H-DPD | Hash-based DPD |
| HAV | Hash assist value |
| FIB | Forwarding Information Base |
| TLV | type-length-value encoding |
| DoS | Denial of Service |
| SMF Router | A MANET Router implementing the protocol specified |
| | in this document |
| SMF Routing | A MANET Routing Domain wherein the protocol |
| Domain | specified in this document is operating |
+--------------+----------------------------------------------------+
3. Applicability Statement
Within dynamic wireless routing topologies, maintaining traditional
forwarding trees to support a multicast routing protocol is often not
as effective as in wired networks due to the reduced reliability and
increased dynamics of mesh topologies [MGL04][GM99]. A basic packet
forwarding service reaching all connected routers running the SMF
protocol within a MANET routing domain may provide a useful group
communication paradigm for various classes of applications, for
example, multimedia streaming, interactive group-based messaging and
applications, peer-to-peer middleware multicasting, and multi-hop
mobile discovery or registration services. SMF is likely only
appropriate for deployment in limited dynamic MANET routing domains
(further defined as administratively scoped multicast forwarding
domains in Section 9.2) so that the flooding process can be
contained.
A design goal is that hosts may also participate in multicast traffic
transmission and reception with standard IP network-layer semantics
(e.g., special or unnecessary encapsulation of IP packets should be
avoided in this case). SMF deployments are able to connect and
interoperate with existing standard multicast protocols operating
within more conventional Internet infrastructures. To this end, a
multicast border router or proxy mechanism MUST be used when deployed
alongside more fixed-infrastructure IP multicast routing such
Protocol Independent Multicast (PIM) variants [RFC3973][RFC4601].
Experimental SMF implementations and deployments have demonstrated
gateway functionality at MANET border routers operating with existing
external IP multicast routing protocols [CDHM07][DHS08][DHG09]. SMF
may be extended or combined with other mechanisms to provide
increased reliability and group-specific filtering; the details for
this are out of the scope of this document.
This document does not presently support forwarding of packets with
directed broadcast addresses as a destination [RFC2644].
4. Overview and Functioning
Figure 1 provides an overview of the logical SMF router architecture,
consisting of "Neighborhood Discovery", "Relay Set Selection", and
"Forwarding Process" components. Typically, relay set selection (or
self-election) occurs based on dynamic input from a neighborhood
discovery process. SMF supports the case where neighborhood
discovery and/or relay set selection information is obtained from a
coexistent process (e.g., a lower-layer mechanism or a unicast
routing protocol using relay sets). In some algorithm designs, the
forwarding decision for a packet can also depend on previous hop or
incoming interface information. The asterisks (*) in Figure 1 mark
the primitives and relationships needed by relay set algorithms
requiring previous-hop packet-forwarding knowledge.
______________ _____________
| | | |
| Neighborhood | | Relay Set |
| Discovery |------------->| Selection |
| | neighbor | |
|______________| info |_____________|
\ /
\ /
neighbor\ /forwarding
info* \ ____________ / status
\ | | /
`-->| Forwarding |<--'
| Process |
~~~~~~~~~~~~~~~~~>|____________|~~~~~~~~~~~~~~~~~>
incoming packet, forwarded packets
interface id*, and
previous hop*
Figure 1: SMF Router Architecture
Certain IP multicast packets, defined in Sections 9.2 and 5, are
"non-forwardable". These multicast packets MUST be ignored by the
SMF forwarding engine. The SMF forwarding engine MAY also work with
policies and management interfaces to allow additional filtering
control over which multicast packets are considered for potential SMF
forwarding. This interface would allow more refined dynamic
forwarding control once such techniques are matured for MANET
operation. At present, further discussion of dynamic control is left
to future work.
Interoperable SMF implementations MUST use compatible DPD approaches
and be able to process the header options defined in this document
for IPv6 operation.
Classic Flooding (CF) is defined as the simplest case of SMF
multicast forwarding. With CF, all SMF routers forward each received
multicast packet exactly once. In this case, the need for any relay
set selection or neighborhood topology information is eliminated, at
the expense of additional network overhead incurred from unnecessary
packet retransmissions. With CF, the SMF DPD functionality is still
required. While SMF supports CF as a mode of operation, the use of
more efficient relay set modes is RECOMMENDED in order to reduce
contention and congestion caused by unnecessary packet
retransmissions [NTSC99].
An efficient reduced relay set is constructed by selecting and
updating, as needed, a subset of all possible routers in a MANET
routing domain to act as SMF forwarders. Known distributed relay set
selection algorithms have demonstrated the ability to provide and
maintain a dynamic connected set for forwarding multicast IP packets
[MDC04]. A few such relay set selection algorithms are described in
the appendices of this document, and the basic designs borrow
directly from previously documented IETF work. SMF relay set
configuration is extensible, and additional relay set algorithms
beyond those specified here can be accommodated in future work.
Determining and maintaining an optimized set of relays generally
requires dynamic neighborhood topology information. Neighborhood
topology discovery functions MAY be provided by a MANET unicast
routing protocol or by using the MANET Neighborhood Discovery
Protocol (NHDP) [RFC6130], operating concurrently with SMF. This
specification also allows alternative lower-layer interfaces (e.g.,
radio router interfaces) to provide the necessary neighborhood
information to aid in supporting more effective relay set selection.
An SMF implementation SHOULD provide the ability for multicast
forwarding state to be dynamically managed per operating network
interface. The relay state maintenance options and interactions are
outlined in Section 7. This document states specific requirements
for neighborhood discovery with respect to the forwarding process and
the relay set selection algorithms described herein. For determining
dynamic relay sets in the absence of other control protocols, SMF
relies on NHDP to assist in IP-layer 2-hop neighborhood discovery and
maintenance for relay set selection. "SMF_TYPE" and "SMF_NBR_TYPE"
Message and Address Block TLV structures (per [RFC5444]) are defined
by this document for use with NHDP to carry SMF-specific information.
It is RECOMMENDED that all routers performing SMF operation in
conjunction with NHDP include these TLV types in any NHDP HELLO
messages generated. This capability allows for routers participating
in SMF to be explicitly identified along with their respective
dynamic relay set algorithm configuration.
5. SMF Packet Processing and Forwarding
The SMF packet processing and forwarding actions are conducted with
the following packet handling activities:
1. Processing of outbound, locally generated multicast packets.
2. Reception and processing of inbound packets on specific network
interfaces.
The purpose of intercepting outbound, locally generated multicast
packets is to apply any added packet marking needed to satisfy the
DPD requirements so that proper forwarding may be conducted. Note
that for some system configurations, the interception of outbound
packets for this purpose is not necessary.
Inbound multicast packets are received by the SMF implementation and
processed for possible forwarding. SMF implementations MUST be
capable of forwarding IP multicast packets with destination addresses
that are not router-local and link-local for IPv6, as defined in
[RFC4291], and that are not within the local network control block as
defined by [RFC5771].
This will support generic multi-hop multicast application needs or
distribute designated multicast traffic ingressing the SMF routing
domain via border routers. The multicast addresses to be forwarded
should be maintained by an a priori list or a dynamic forwarding
information base (FIB) that MAY interact with future MANET dynamic
group membership extensions or management functions.
The SL-MANET-ROUTERS multicast group is defined to contain all
routers within an SMF routing domain, so that packets transmitted to
the multicast address associated with the group will be attempted to
be delivered to all connected routers running SMF. Due to the mobile
nature of a MANET, routers running SMF may not be topologically
connected at particular times. For IPv6, SL-MANET-ROUTERS is
specified to be "site-local". Minimally, SMF MUST forward, as
instructed by the relay set selection algorithm, unique (non-
duplicate) packets received for SL-MANET-ROUTERS when the Time to
Live (TTL) / hop limit or hop limit value in the IP header is greater
than 1. SMF MUST forward all additional global-scope multicast
addresses specified within the dynamic FIB or configured list as
well. In all cases, the following rules MUST be observed for SMF
multicast forwarding:
1. Any IP packets not addressed to an IP multicast address MUST NOT
be forwarded by the SMF forwarding engine.
2. IP multicast packets with TTL/hop limit <= 1 MUST NOT be
forwarded.
3. Link local IP multicast packets MUST NOT be forwarded.
4. Incoming IP multicast packets with an IP source address matching
one of those of the local SMF router interface(s) MUST NOT be
forwarded.
5. Received frames with the Media Access Control (MAC) source
address matching any MAC address of the router's interfaces MUST
NOT be forwarded.
6. Received packets for which SMF cannot reasonably ensure temporal
DPD uniqueness MUST NOT be forwarded.
7. Prior to being forwarded, the TTL/hop limit of the forwarded
packet MUST be decremented by one.
Note that rule #4 is important because over some types of wireless
interfaces, the originating SMF router may receive retransmissions of
its own packets when they are forwarded by adjacent routers. This
rule avoids unnecessary retransmission of locally generated packets
even when other forwarding decision rules would apply.
An additional processing rule also needs to be considered based upon
a potential security threat. As discussed in Section 10, there is a
potential DoS attack that can be conducted by remotely "previewing"
(e.g., via a directional receive antenna) packets that an SMF router
would be forwarding and conducting a "pre-play" attack by
transmitting the packet before the SMF router would otherwise receive
it, but with a reduced TTL/hop limit field value. This form of
attack can cause an SMF router to create a DPD entry that would block
the proper forwarding of the valid packet (with correct TTL/hop
limit) through the SMF routing domain. A RECOMMENDED approach to
prevent this attack, when it is a concern, would be to cache temporal
packet TTL/hop limit values along with the per-packet DPD state (hash
value(s) and/or identifier as described in Section 6). Then, if a
subsequent matching (with respect to DPD) packet arrives with a
larger TTL/hop limit value than the packet that was previously
forwarded, SMF should forward the new packet and update the TTL/hop
limit value cached with corresponding DPD state to the new, larger
TTL/hop limit value. There may be temporal cases where SMF would
unnecessarily forward some duplicate packets using this approach, but
those cases are expected to be minimal and acceptable when compared
with the potential threat of denied service.
Once the SMF multicast forwarding rules have been applied, an SMF
implementation MUST make a forwarding decision dependent upon the
relay set selection algorithm in use. If the SMF implementation is
using Classic Flooding (CF), the forwarding decision is implicit once
DPD uniqueness is determined. Otherwise, a forwarding decision
depends upon the current interface-specific relay set state. The
descriptions of the relay set selection algorithms in the appendices
to this document specify the respective heuristics for multicast
packet forwarding and specific DPD or other processing required to
achieve correct SMF behavior in each case. For example, one class of
forwarding is based upon relay set selection status and the packet's
previous hop, while other classes designate the local SMF router as a
forwarder for all neighboring routers.
6. SMF Duplicate Packet Detection
Duplicate packet detection (DPD) is often a requirement in MANET or
wireless mesh packet forwarding mechanisms because packets may be
transmitted out via the same physical interface as the one over which
they were received. Routers may also receive multiple copies of the
same packets from different neighbors or interfaces. SMF operation
requires DPD, and implementations MUST provide mechanisms to detect
and reduce the likelihood of forwarding duplicate multicast packets
using temporal packet identification. It is RECOMMENDED this be
implemented by keeping a history of recently processed multicast
packets for comparison with incoming packets. A DPD packet cache
history SHOULD be kept long enough so as to span the maximum network
traversal lifetime, MAX_PACKET_LIFETIME, of multicast packets being
forwarded within an SMF routing domain. The DPD mechanism SHOULD
avoid keeping unnecessary state for packet flows such as those that
are locally generated or link-local destinations that would not be
considered for forwarding, as presented in Section 5.
For both IPv4 and IPv6, this document describes two basic multicast
duplicate packet detection mechanisms: header content identification-
based (I-DPD) and hash-based (H-DPD) duplicate packet detection.
I-DPD is a mechanism using specific packet headers, and option
headers in the case of IPv6, in combination with flow state to
estimate the temporal uniqueness of a packet. H-DPD uses hashing
over header fields and payload of a multicast packet to provide an
estimation of temporal uniqueness.
Trade-offs of the two approaches to DPD merit different
considerations dependent upon the specific SMF deployment scenario.
Because of the potential addition of a hop-by-hop option header with
IPv6, all SMF routers in the same SMF deployments MUST be configured
so as to use a common mechanism and DPD algorithm. The main
difference between IPv4 and IPv6 SMF DPD specifications is the
avoidance of any additional header options for IPv4.
For each network interface, SMF implementations MUST maintain DPD
packet state as needed to support the forwarding heuristics of the
relay set algorithm used. In general, this involves keeping track of
previously forwarded packets so that duplicates are not forwarded,
but some relay techniques have additional considerations, such as
those discussed in Appendix B.2.
Additional details of I-DPD and H-DPD processing and maintenance for
different classes of packets are described in the following
subsections.
6.1. IPv6 Duplicate Packet Detection
This section describes the mechanisms and options for SMF IPv6 DPD.
The base IPv6 packet header does not provide an explicit packet
identification header field that can be exploited for I-DPD. The
following options are therefore described to support IPv6 DPD:
1. a hop-by-hop SMF_DPD option header, defined in this document
(Section 6.1.1),
2. the use of IPv6 fragment header fields for I-DPD, if one is
present (Section 6.1.2),
3. the use of IPsec sequencing for I-DPD when a non-fragmented,
IPsec header is detected (Section 6.1.2), and
4. an H-DPD approach assisted, as needed, by the SMF_DPD option
header (Section 6.1.3).
SMF MUST provide a DPD marking module that can insert the hop-by-hop
IPv6 header option, defined in Section 6.1.1. This module MUST be
invoked after any source-based fragmentation that may occur with
IPv6, so as to ensure that all fragments are suitably marked. SMF
IPv6 DPD is presently specified to allow either a packet hash or
header identification method for DPD. An SMF implementation MUST be
configured to operate either in I-DPD or H-DPD mode and perform the
corresponding tasks, outlined in Sections 6.1.2 and 6.1.3.
6.1.1. IPv6 SMF_DPD Option Header
This section defines an IPv6 Hop-by-Hop Option [RFC2460], SMF_DPD, to
serve the purpose of unique packet identification for IPv6 I-DPD.
Additionally, the SMF_DPD option header provides a mechanism to
guarantee non-collision of hash values for different packets when
H-DPD is used.
If this is the only hop-by-hop option present, the optional TaggerId
field (see below) is not included, and the size of the DPD packet
identifier (sequence number) or hash token is 24 bits or less, this
will result in the addition of 8 bytes to the IPv6 packet header
including the "Next Header", "Header Extension Length", SMF_DPD
option fields, and padding.
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... |0|0|0| 01000 | Opt. Data Len |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|H| DPD Identifier Option Fields or Hash Assist Value ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 2: IPv6 SMF_DPD Hop-by-Hop Option Header
"Option Type" = 00001000. The highest order three bits are 000
because this specification requires that routers not recognizing this
option type skip over this option and continue processing the header
and that the option must not change en route [RFC2460].
"Opt. Data Len" = Length of option content (i.e., 1 + (<IdType> ?
(<IdLen> + 1): 0) + Length(DPD ID)).
"H-bit" = a hash indicator bit value identifying DPD marking type. 0
== sequence-based approach with optional TaggerId and a tuple-based
sequence number. 1 == indicates a hash assist value (HAV) field
follows to aid in avoiding hash-based DPD collisions.
When the "H-bit" is cleared (zero value), the SMF_DPD format to
support I-DPD operation is specified as shown in Figure 3 and defines
the extension header in accordance with [RFC2460].
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... |0|0|0| 01000 | Opt. Data Len |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|0|TidTy| TidLen| TaggerId (optional) ... |
+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | Identifier ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 3: IPv6 SMF_DPD Option Header in I-DPD mode
"TidTy" = a 3-bit field indicating the presence and type of the
optional TaggerId field.
"TidLen" = a 4-bit field indicating the length (in octets) of the
following TaggerId field.
"TaggerId" = a field, is used to differentiate multiple ingressing
border gateways that may commonly apply the SMF_DPD option header to
packets from a particular source. Table 1 lists the TaggerId types
used in this document:
+---------+---------------------------------------------------------+
| Name | Purpose |
+---------+---------------------------------------------------------+
| NULL | Indicates no TaggerId field is present. "TidLen" MUST |
| | also be set to ZERO. |
| DEFAULT | A TaggerId of non-specific context is present. "TidLen |
| | + 1" defines the length of the TaggerId field in bytes. |
| IPv4 | A TaggerId representing an IPv4 address is present. The |
| | "TidLen" MUST be set to 3. |
| IPv6 | A TaggerId representing an IPv6 address is present. The |
| | "TidLen" MUST be set to 15. |
+---------+---------------------------------------------------------+
Table 1: TaggerId Types
This format allows a quick check of the "TidTy" field to determine if
a TaggerId field is present. If "TidTy" is NULL, then the length of
the DPD packet <Identifier> field corresponds to (<Opt. Data Len> -
1). If the <TidTy> is non-NULL, then the length of the TaggerId
field is equal to (<TidLen> - 1), and the remainder of the option
data comprises the DPD packet <Identifier> field. When the TaggerId
field is present, the <Identifier> field can be considered a unique
packet identifier in the context of the <TaggerId:srcAddr:dstAddr>
tuple. When the TaggerId field is not present, then it is assumed
that the source applied the SMF_DPD option and the <Identifier> can
be considered unique in the context of the IPv6 packet header
<srcAddr:dstAddr> tuple. IPv6 I-DPD operation details are in
Section 6.1.2.
When the "H-bit" in the SMF_DPD option data is set, the data content
value is interpreted as a hash assist value (HAV) used to facilitate
H-DPD operation. In this case, the source or ingressing gateways
apply the SMF_DPD with an HAV only when required to differentiate the
hash value of a new packet with respect to hash values in the DPD
cache. This situation can be detected locally on the router by
running the hash algorithm and checking the DPD cache, prior to
ingressing a previously unmarked packet or a locally sourced packet.
This helps to guarantee the uniqueness of generated hash values when
H-DPD is used. Additionally, this avoids the added overhead of
applying the SMF_DPD option header to every packet. For many hash
algorithms, it is expected that only sparse use of the SMF_DPD option
may be required. The format of the SMF_DPD option header for H-DPD
operation is given in Figure 4.
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... |0|0|0| OptType | Opt. Data Len |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|1| Hash Assist Value (HAV) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 4: IPv6 SMF_DPD Option Header in H-DPD Mode
The SMF_DPD option should be applied with an HAV to produce a unique
hash digest for packets within the context of the IPv6 packet header
<srcAddr>. The size of the HAV field is implied by "Opt. Data Len".
The appropriate size of the field depends upon the collision
properties of the specific hash algorithm used. More details on IPv6
H-DPD operation are provided in Section 6.1.3.
6.1.2. IPv6 Identification-Based DPD
Table 2 summarizes the IPv6 I-DPD processing and forwarding decision
approach. Within the table, '*' indicates an ignore field condition.
+-------------+-----------+-----------+-----------------------------+
| IPv6 | IPv6 | IPv6 | SMF IPv6 I-DPD Mode Action |
| Fragment | IPsec | I-DPD | |
| Header | Header | Header | |
+-------------+-----------+-----------+-----------------------------+
| Present | * | Not | Use Fragment Header I-DPD |
| | | Present | Check and Process for |
| | | | Forwarding |
| Not Present | Present | Not | Use IPsec Header I-DPD |
| | | Present | Check and Process for |
| | | | Forwarding |
| Present | * | Present | Invalid; do not forward. |
| Not Present | Present | Present | Invalid; do not forward. |
| Not Present | Not | Not | Add I-DPD Header, and |
| | Present | Present | Process for Forwarding |
| Not Present | Not | Present | Use I-DPD Header Check and |
| | Present | | Process for Forwarding |
+-------------+-----------+-----------+-----------------------------+
Table 2: IPv6 I-DPD Processing Rules
1. If a received IPv6 multicast packet is an IPv6 fragment, SMF MUST
use the fragment extension header fields for packet
identification. This identifier can be considered unique in the
context of the <srcAddr:dstAddr> of the IP packet.
2. If the packet is an unfragmented IPv6 IPsec packet, SMF MUST use
IPsec fields for packet identification. The IPsec header
<sequence> field can be considered a unique identifier in the
context of the <IPsecType:srcAddr:dstAddr:SPI> where "IPsecType"
is either Authentication Header (AH) or Encapsulating Security
Payload (ESP) [RFC4302].
3. For unfragmented, non-IPsec IPv6 packets, the use of the SMF_DPD
option header is necessary to support I-DPD operation. The
SMF_DPD option header is applied in the context of the <srcAddr>
of the IP packet. Hosts or ingressing SMF gateways are
responsible for applying this option to support DPD. Table 3
summarizes these packet identification types:
+-----------+---------------------------------+---------------------+
| IPv6 | Packet DPD ID Context | Packet DPD ID |
| Packet | | |
| Type | | |
+-----------+---------------------------------+---------------------+
| Fragment | <srcAddr:dstAddr> | <fragmentOffset:id> |
| IPsec | <IPsecType:srcAddr:dstAddr:SPI> | <sequence> |
| Packet | | |
| Regular | <[TaggerId:]srcAddr:dstAddr> | <SMF_DPD option |
| Packet | | header id> |
+-----------+---------------------------------+---------------------+
Table 3: IPv6 I-DPD Packet Identification Types
"IPsecType" is either Authentication Header (AH) or Encapsulating
Security Payload (ESP).
The "TaggerId" is an optional field of the IPv6 SMF_DPD option
header.
6.1.3. IPv6 Hash-Based DPD
A default hash-based DPD approach (H-DPD) for use by SMF is specified
as follows. An SHA-1 [RFC3174] hash of the non-mutable header
fields, options fields, and data content of the IPv6 multicast packet
is used to produce a 160-bit digest. The approach for calculating
this hash value SHOULD follow the same guidelines described for
calculating the Integrity Check Value (ICV) described in [RFC4302]
with respect to non-mutable fields. This approach should have a
reasonably low probability of digest collision when packet headers
and content are varying. SHA-1 is being applied in SMF only to
provide a low probability of collision and is not being used for
cryptographic or authentication purposes. A history of the packet
hash values SHOULD be maintained within the context of the IPv6
packet header <srcAddr>. SMF ingress points (i.e., source hosts or
gateways) use this history to confirm that new packets are unique
with respect to their hash value. The hash assist value (HAV) field
described in Section 6.1.1 is provided as a differentiating field
when a digest collision would otherwise occur. Note that the HAV is
an immutable option field, and SMF MUST process any included HAV
values (see Section 6.1.1) in its hash calculation.
If a packet results in a digest collision (i.e., by checking the
H-DPD digest history) within the DPD cache kept by SMF forwarders,
the packet SHOULD be silently dropped. If a digest collision is
detected at an SMF ingress point, the H-DPD option header is
constructed with a randomly generated HAV. An HAV is recalculated as
needed to produce a non-colliding hash value prior to forwarding.
The multicast packet is then forwarded with the added IPv6 SMF_DPD
option header. A common hash approach MUST be used by SMF routers
for the applied HAV to consistently avoid hash collision and thus
inadvertent packet drops.
The SHA-1 indexing and IPv6 HAV approaches are specified at present
for consistency and robustness to suit experimental uses. Future
approaches and experimentation may discover design trade-offs in hash
robustness and efficiency worth considering. Enhancements MAY
include reducing the maximum payload length that is processed,
determining shorter indexes, or applying more efficient hashing
algorithms. Use of the HAV functionality may allow for application
of "lighter-weight" hashing techniques that might not have been
initially considered otherwise due to poor collision properties.
Such techniques could reduce packet-processing overhead and memory
requirements.
6.2. IPv4 Duplicate Packet Detection
This section describes the mechanisms and options for IPv4 DPD. The
following areas are described to support IPv4 DPD:
1. the use of IPv4 fragment header fields for I-DPD when they exist
(Section 6.2.1),
2. the use of IPsec sequencing for I-DPD when a non-fragmented IPv4
IPsec packet is detected (Section 6.2.1), and
3. an H-DPD approach(Section 6.2.2) when neither of the above cases
can be applied.
Although the IPv4 datagram has a 16-bit Identification (ID) field as
specified in [RFC0791], it cannot be relied upon for DPD purposes due
to common computer operating system implementation practices and the
reasons described in the updated specification of the IPv4 ID Field
[IPV4-ID-UPDATE]. An SMF IPv4 DPD marking option like the IPv6
SMF_DPD option header is not specified since IPv4 header options are
not as tractable for hosts as they are for IPv6. However, when IPsec
is applied or IPv4 packets have been fragmented, the I-DPD approach
can be applied reliably using the corresponding packet identifier
fields described in Section 6.2.1. For the general IPv4 case (non-
IPsec and non-fragmented packets), the H-DPD approach of
Section 6.2.2 is RECOMMENDED.
Since IPv4 SMF does not specify an option header, the
interoperability constraints are looser than in the IPv6 version, and
forwarders may operate with mixed H-DPD and I-DPD modes as long as
they consistently perform the appropriate DPD routines outlined in
the following sections. However, it is RECOMMENDED that a deployment
be configured with a common mode for operational consistency.
6.2.1. IPv4 Identification-Based DPD
Table 4 summarizes the IPv4 I-DPD processing approach once a packet
has passed the basic forwardable criteria described in Section 5. To
summarize, for IPv4, I-DPD is applicable only for packets that have
been fragmented or have IPsec applied. In Table 4, '*' indicates an
ignore field condition. DF, MF, and Fragment offset correspond to
related fields and flags defined in [RFC0791].
+------+------+----------+---------+--------------------------------+
| DF | MF | Fragment | IPsec | IPv4 I-DPD Action |
| flag | flag | offset | | |
+------+------+----------+---------+--------------------------------+
| 1 | 1 | * | * | Invalid; do not forward. |
| 1 | 0 | nonzero | * | Invalid; do not forward. |
| * | 0 | zero | not | Use H-DPD check instead |
| | | | Present | |
| * | 0 | zero | Present | IPsec enhanced Tuple I-DPD |
| | | | | Check and Process for |
| | | | | Forwarding |
| 0 | 0 | nonzero | * | Extended Fragment Offset Tuple |
| | | | | I-DPD Check and Process for |
| | | | | Forwarding |
| 0 | 1 | zero or | * | Extended Fragment Offset Tuple |
| | | nonzero | | I-DPD Check and Process for |
| | | | | Forwarding |
+------+------+----------+---------+--------------------------------+
Table 4: IPv4 I-DPD Processing Rules
For performance reasons, IPv4 network fragmentation and reassembly of
multicast packets within wireless MANET networks should be minimized,
yet SMF provides the forwarding of fragments when they occur. If the
IPv4 multicast packet is a fragment, SMF MUST use the fragmentation
header fields for packet identification. This identification can be
considered temporally unique in the context of the <protocol:srcAddr:
dstAddr> of the IPv4 packet. If the packet is an unfragmented IPv4
IPsec packet, SMF MUST use IPsec fields for packet identification.
The IPsec header <sequence> field can be considered a unique
identifier in the context of the <IPsecType:srcAddr:dstAddr:SPI>
where "IPsecType" is either AH or ESP [RFC4302]. Table 5 summarizes
these packet identification types:
+-----------+---------------------------------+---------------------+
| IPv4 | Packet Identification Context | Packet Identifier |
| Packet | | |
| Type | | |
+-----------+---------------------------------+---------------------+
| Fragment | <protocol:srcAddr:dstAddr> | <fragmentOffset:id> |
| IPsec | <IPsecType:srcAddr:dstAddr:SPI> | <sequence> |
| Packet | | |
+-----------+---------------------------------+---------------------+
Table 5: IPv4 I-DPD Packet Identification Types
"IPsecType" is either Authentication Header (AH) or Encapsulating
Security Payload (ESP).
6.2.2. IPv4 Hash-Based DPD
The hashing technique here is similar to that specified for IPv6 in
Section 6.1.3, but the H-DPD header option with HAV is not
considered. To ensure consistent IPv4 H-DPD operation among SMF
routers, a default hashing approach is specified. A common DPD
hashing algorithm for an SMF routing area is RECOMMENDED because
colliding hash values for different packets result in "false
positive" duplicate packet detection, and there is small probability
that valid packets may be dropped based on the hashing technique
used. Since the "hash assist value" approach is not available for
IPv4, use of a common hashing approach minimizes the probability of
hash collision packet drops over multiple hops of forwarding.
SMF MUST perform a SHA-1 [RFC3174] hash of the immutable header
fields, option fields, and data content of the IPv4 multicast packet
resulting in a 160-bit digest. The approach for calculating the hash
value SHOULD follow the same guidelines described for calculating the
Integrity Check Value (ICV) described in [RFC4302] with respect to
non-mutable fields. A history of the packet hash values SHOULD be
maintained in the context of <protocol:srcAddr:dstAddr>. The context
for IPv4 is more specific than that of IPv6 since the SMF_DPD HAV
cannot be employed to mitigate hash collisions. A RECOMMENDED
implementation detail for IPv4 H-DPD is to concatenate the 16-bit
IPv4 ID value with the computed hash value as an extended DPD hash
value that may provide reduced hash collisions in the cases where the
IPv4 ID field is being set by host operating systems or gateways.
When this approach is taken, the use of the supplemental "internal
hash" technique as described in Section 10 is RECOMMENDED as a
security measure.
The SHA-1 hash is specified at present for consistency and
robustness. Future approaches and experimentation may discover
design trade-offs in hash robustness and efficiency worth considering
for future revisions of SMF. This MAY include reducing the packet
payload length that is processed, determining shorter indexes, or
applying a more efficient hashing algorithm.
7. Relay Set Selection
SMF is flexible in its support of different reduced relay set
mechanisms for efficient flooding, the constraints imposed herein
being detailed in this section.
7.1. Non-Reduced Relay Set Forwarding
SMF implementations MUST support CF as a basic forwarding mechanism
when reduced relay set information is not available or not selected
for operation. In CF mode, each router transmits a packet once that
has passed the SMF forwarding rules. The DPD techniques described in
Section 6 are critical to proper operation and prevention of
duplicate packet retransmissions by the same relays.
7.2. Reduced Relay Set Forwarding
MANET reduced relay sets are often achieved by distributed algorithms
that can dynamically calculate a topological connected dominating set
(CDS).
A goal of SMF is to apply reduced relay sets for more efficient
multicast dissemination within dynamic topologies. To accomplish
this, an SMF implementation MUST support the ability to modify its
multicast packet forwarding rules based upon relay set state received
dynamically during operation. In this way, SMF operates effectively
as neighbor adjacencies or multicast forwarding policies within the
topology change.
In early SMF experimental prototyping, the relay set information was
derived from coexistent unicast routing control plane traffic
flooding processes [MDC04]. From this experience, extra pruning
considerations were sometimes required when utilizing a relay set
from a separate routing protocol process. As an example, relay sets
formed for the unicast control plane flooding MAY include additional
redundancy that may not be desired for multicast forwarding use
(e.g., biconnected relay set).
Here is a recommended criteria list for SMF relay set selection
algorithm candidates:
1. Robustness to topological dynamics and mobility
2. Localized election or coordination of any relay sets
3. Reasonable minimization of CDS relay set size given the above
constraints
4. Heuristic support for preference or election metrics
Some relay set algorithms meeting these criteria are described in the
appendices of this document. Additional relay set selection
algorithms may be specified in separate specifications in the future.
Each appendix subsection in this document can serve as a template for
specifying additional relay algorithms.
Figure 5 depicts an information flow diagram of possible relay set
control options. The SMF Relay Set State represents the information
base that is used by SMF in the forwarding decision process. The
diagram demonstrates that the SMF Relay Set State may be determined
by three fundamentally different methods:
o Independent operation with NHDP [RFC6130] input providing dynamic
network neighborhood adjacency information, used by a particular
relay set selection algorithm.
o Slave operation with an existing unicast MANET routing protocol,
capable of providing CDS election information for use by SMF.
o Cross-layer operation that may involve L2 triggers or information
describing neighbors or links.
Other heuristics to influence and control election can come from
network management or other interfaces as shown on the right of
Figure 5. CF mode simplifies the control and does not require other
input but relies solely on DPD.
Possible L2 Trigger/Information
|
|
______________ ______v_____ __________________
| MANET | | | | |
| Neighborhood | | Relay Set | | Other Heuristics |
| Discovery |----------->| Selection |<------|(Preference, etc.)|
| Protocol | neighbor | Algorithm | | Net Management |
|______________| info |____________| |__________________|
\ /
\ /
neighbor\ / Dynamic Relay
info* \ ____________ / Set Status
\ | SMF | / (State, {neighbor info})
`-->| Relay Set |<--'
| State |
-->|____________|
/
/
______________
| Coexistent |
| MANET |
| Unicast |
| Process |
|______________|
Figure 5: SMF Reduced Relay Set Information Flow
Following is further discussion of the three styles of SMF operation
with reduced relay sets as illustrated in Figure 5:
1. Independent operation: In this case, SMF operates independently
from any unicast routing protocols. To support reduced relay
sets, SMF MUST perform its own relay set selection using
information gathered from signaling. It is RECOMMENDED that an
associated NHDP process be used for this signaling. NHDP
messaging SHOULD be appended with additional [RFC5444] type-
length-value (TLV) content as to support SMF-specific
requirements as discussed in [RFC6130] and to support specific
relay set operation as described in the appendices of this
document or future specifications. Unicast routing protocols may
coexist, even using the same NHDP process, but signaling that
supports reduced relay set selection for SMF is independent of
these protocols.
2. Operation with CDS-aware unicast routing protocol: In this case,
a coexistent unicast routing protocol provides dynamic relay set
state based upon its own control plane CDS or neighborhood
discovery information.
3. Cross-layer operation: In this case, SMF operates using
neighborhood status and triggers from a cross-layer information
base for dynamic relay set selection and maintenance (e.g.,
lower-link layer).
8. SMF Neighborhood Discovery Requirements
This section defines the requirements for use of the MANET
Neighborhood Discovery Protocol (NHDP) [RFC6130] to support SMF
operation. Note that basic CF forwarding requires no neighborhood
topology knowledge since in this configured mode, every SMF router
relays all traffic. Supporting more reduced SMF relay set operation
requires the discovery and maintenance of dynamic neighborhood
topology information. NHDP can be used to provide this necessary
information; however, there are SMF-specific requirements for NHDP
use. This is the case for both "independent" SMF operation where
NHDP is being used specifically to support SMF or when one NHDP
instance is used for both SMF and a coexistent MANET unicast routing
protocol.
NHDP HELLO messages and the resultant neighborhood information base
are described separately within the NHDP specification. To
summarize, NHDP provides the following basic functions:
1. 1-hop neighbor link sensing and bidirectionality checks of
neighbor links,
2. 2-hop neighborhood discovery including collection of 2-hop
neighbors and connectivity information,
3. Collection and maintenance of the above information across
multiple interfaces, and
4. A method for signaling SMF information throughout the 2-hop
neighborhood through the use of TLV extensions.
Appendices A-C of this document describe CDS-based relay set
selection algorithms that can achieve efficient SMF operation, even
in dynamic, mobile networks and each of the algorithms has been
initially experimented with in a working SMF prototype [MDDA07].
When using these algorithms in conjunction with NHDP, a method
verifying neighbor SMF operation is required in order to ensure
correct relay set selection. NHDP, along with SMF operation
verification, provides the necessary information required by these
algorithms to conduct relay set selection. Verification of SMF
operation may be done administratively or through the use of the SMF
relay algorithms TLVs defined in the following subsections. Use of
the SMF relay algorithm TLVs is RECOMMENDED when using NHDP for SMF
neighborhood discovery.
Section 8.1 specifies SMF-specific TLV types, supporting general SMF
operation or supporting the algorithms described in the appendices.
The appendices describing several relay set algorithms also specify
any additional requirements for use with NHDP and reference the
applicable TLV types as needed.
8.1. SMF Relay Algorithm TLV Types
This section specifies TLV types to be used within NHDP messages to
identify the CDS relay set selection algorithm(s) in use. Two TLV
types are defined: one Message TLV type and one Address Block TLV
type.
8.1.1. SMF Message TLV Type
The Message TLV type denoted SMF_TYPE is used to identify the
existence of an SMF instance operating in conjunction with NHDP.
This Message TLV type makes use of the extended type field as defined
by [RFC5444] to convey the CDS relay set selection algorithm
currently in use by the SMF message originator. When NHDP is used to
support SMF operation, the SMF_TYPE TLV, containing the extended type
field with the appropriate value, SHOULD be included in NHDP_HELLO
messages (HELLO messages as defined in [RFC6130]). This allows SMF
routers to learn when neighbors are configured to use NHDP for
information exchange including algorithm type and related algorithm
information. This information can be used to take action, such as
ignoring neighbor information using incompatible algorithms. It is
possible that SMF neighbors MAY be configured differently and still
operate cooperatively, but these cases will vary dependent upon the
algorithm types designated.
This document defines a Message TLV type as specified in Table 6
conforming to [RFC5444]. The TLV extended type field is used to
contain the sender's "Relay Algorithm Type". The interpretation of
the "value" content of these TLVs is defined per "Relay Algorithm
Type" and may contain algorithm-specific information.
+---------------+----------------+--------------------+
| | TLV Syntax | Field Values |
+---------------+----------------+--------------------+
| type | <tlv-type> | SMF_TYPE |
| extended type | <tlv-type-ext> | <relayAlgorithmId> |
| length | <length> | variable |
| value | <value> | variable |
+---------------+----------------+--------------------+
Table 6: SMF Type Message TLV
In Table 6, <relayAlgorithmId> is an 8-bit field containing a number
0-255 representing the "Relay Algorithm Type" of the originator
address of the corresponding NHDP message.
Values for the <relayAlgorithmId> are defined in Table 7. The table
provides value assignments, future IANA assignment spaces, and an
experimental space. The experimental space use MUST NOT assume
uniqueness; thus, it SHOULD NOT be used for general interoperable
deployment prior to official IANA assignment.
+-------------+--------------------+--------------------------------+
| Type Value | Extended Type | Algorithm |
| | Value | |
+-------------+--------------------+--------------------------------+
| SMF_TYPE | 0 | CF |
| SMF_TYPE | 1 | S-MPR |
| SMF_TYPE | 2 | E-CDS |
| SMF_TYPE | 3 | MPR-CDS |
| SMF_TYPE | 4-127 | Future Assignment STD action |
| SMF_TYPE | 128-239 | No STD action required |
| SMF_TYPE | 240-255 | Experimental Space |
+-------------+--------------------+--------------------------------+
Table 7: SMF Relay Algorithm Type Values
Acceptable <length> and <value> fields of an SMF_TYPE TLV are
dependent on the extended type value (i.e., relay algorithm type).
The appropriate algorithm type, as conveyed in the <tlv-type-ext>
field, defines the meaning and format of its TLV <value> field. For
the algorithms defined by this document, see the appropriate appendix
for the <value> field format.
8.1.2. SMF Address Block TLV Type
An Address Block TLV type, denoted SMF_NBR_TYPE (i.e., SMF neighbor
relay algorithm) is specified in Table 8. This TLV enables CDS relay
algorithm operation and configuration to be shared among 2-hop
neighborhoods. Some relay algorithms require 2-hop neighbor
configuration in order to correctly select relay sets. It is also
useful when mixed relay algorithm operation is possible. Some
examples of mixed use are outlined in the appendices.
The message SMF_TYPE TLV and Address Block SMF_NBR_TYPE TLV types
share a common format.
+---------------+----------------+--------------------+
| | TLV syntax | Field Values |
+---------------+----------------+--------------------+
| type | <tlv-type> | SMF_NBR_TYPE |
| extended type | <tlv-type-ext> | <relayAlgorithmId> |
| length | <length> | variable |
| value | <value> | variable |
+---------------+----------------+--------------------+
Table 8: SMF Type Address Block TLV
<relayAlgorithmId> in Table 8 is an 8-bit unsigned integer field
containing a number 0-255 representing the "Relay Algorithm Type"
value that corresponds to any associated address in the address
block. Note that "Relay Algorithm Type" values for 2-hop neighbors
can be conveyed in a single TLV or multiple value TLVs as described
in [RFC5444]. It is expected that SMF routers using NHDP construct
address blocks with SMF_NBR_TYPE TLVs to advertise "Relay Algorithm
Type" and to advertise neighbor algorithm values received in SMF_TYPE
TLVs from those neighbors.
Again, values for the <relayAlgorithmId> are defined in Table 7.
The interpretation of the "value" field of SMF_NBR_TYPE TLVs is
defined per "Relay Algorithm Type" and may contain algorithm-specific
information. See the appropriate appendix for definitions of value
fields for the algorithms defined by this document.
9. SMF Border Gateway Considerations
It is expected that SMF will be used to provide simple forwarding of
multicast traffic within a MANET or mesh routing topology. A border
router gateway approach should be used to allow interconnection of
SMF routing domains with networks using other multicast routing
protocols, such as PIM. It is important to note that there are many
scenario-specific issues that should be addressed when discussing
border multicast routers. At the present time, experimental
deployments of SMF and PIM border router approaches have been
demonstrated [DHS08]. Some of the functionality border routers may
need to address includes the following:
1. Determination of which multicast group traffic transits the
border router whether entering or exiting the attached SMF
routing domain.
2. Enforcement of TTL/hop limit threshold or other scoping policies.
3. Any marking or labeling to enable DPD on ingressing packets.
4. Interface with exterior multicast routing protocols.
5. Possible operation with multiple border routers (presently beyond
the scope of this document).
6. Provisions for participating non-SMF devices (routers or hosts).
Each of these areas is discussed in more detail in the following
subsections. Note the behavior of SMF border routers is the same as
that of non-border SMF routers when forwarding packets on interfaces
within the SMF routing domain. Packets that are passed outbound to
interfaces operating fixed-infrastructure multicast routing protocols
SHOULD be evaluated for duplicate packet status since present
standard multicast forwarding mechanisms do not usually perform this
function.
9.1. Forwarded Multicast Groups
Mechanisms for dynamically determining groups for forwarding into a
MANET SMF routing domain is an evolving technology area. Ideally,
only traffic for which there is active group membership should be
injected into the SMF domain. This can be accomplished by providing
an IPv4 Internet Group Membership Protocol (IGMP) or IPv6 Multicast
Listener Discovery (MLD) proxy protocol so that MANET SMF routers can
inform attached border routers (and hence multicast networks) of
their current group membership status. For specific systems and
services, it may be possible to statically configure group membership
joins in border routers, but it is RECOMMENDED that some form of
IGMP/MLD proxy or other explicit, dynamic control of membership be
provided. Specification of such an IGMP/MLD proxy protocol is beyond
the scope of this document.
For outbound traffic, SMF border routers perform duplicate packet
detection and forward non-duplicate traffic that meets TTL/hop limit
and scoping criteria to interfaces external to the SMF routing
domain. Appropriate IP multicast routing (e.g., PIM-based solutions)
on those interfaces can make further forwarding decisions with
respect to the multicast packet. Note that the presence of multiple
border routers associated with a MANET routing domain raises
additional issues. This is further discussed in Section 9.4 but
further work is expected to be needed here.
9.2. Multicast Group Scoping
Multicast scoping is used by network administrators to control the
network routing domains reachable by multicast packets. This is
usually done by configuring external interfaces of border routers in
the border of a routing domain to not forward multicast packets that
must be kept within the SMF routing domain. This is commonly done
based on TTL/hop limit of messages or by using administratively
scoped group addresses. These schemes are known respectively as:
1. TTL scoping.
2. Administrative scoping.
For IPv4, network administrators can configure border routers with
the appropriate TTL/hop limit thresholds or administratively scoped
multicast groups for the router interfaces as with any traditional
multicast router. However, for the case of TTL/hop limit scoping, it
SHOULD be taken into account that the packet could traverse multiple
hops within the MANET SMF routing domain before reaching the border
router. Thus, TTL thresholds SHOULD be selected carefully.
For IPv6, multicast address spaces include information about the
scope of the group. Thus, border routers of an SMF routing domain
know if they must forward a packet based on the IPv6 multicast group
address. For the case of IPv6, it is RECOMMENDED that a MANET SMF
routing domain be designated a site-scoped multicast domain. Thus,
all IPv6 site-scoped multicast packets in the range FF05::/16 SHOULD
be kept within the MANET SMF routing domain by border routers. IPv6
packets in any other wider range scopes (i.e., FF08::/16, FF0B::/16,
and FF0E::16) MAY traverse border routers unless other restrictions
different from the scope applies.
Given that scoping of multicast packets is performed at the border
routers and given that existing scoping mechanisms are not designed
to work with mobile routers, it is assumed that non-border routers
running SMF will not stop forwarding multicast data packets of an
appropriate site scoping. That is, it is assumed that an SMF routing
domain is a site-scoped multicast area.
9.3. Interface with Exterior Multicast Routing Protocols
The traditional operation of multicast routing protocols is tightly
integrated with the group membership function. Leaf routers are
configured to periodically gather group membership information, while
intermediate routers conspire to create multicast trees connecting
routers with directly connected multicast sources and routers with
active multicast receivers. In the concrete case of SMF, border
routers can be considered leaf routers. Mechanisms for multicast
sources and receivers to interoperate with border routers over the
multi-hop MANET SMF routing domain as if they were directly connected
to the router need to be defined. The following issues need to be
addressed:
1. A mechanism by which border routers gather membership information
2. A mechanism by which multicast sources are known by the border
router
3. A mechanism for exchange of exterior routing protocol messages
across the SMF routing domain if the SMF routing domain is to
provide transit connectivity for multicast traffic.
It is beyond the scope of this document to address implementation
solutions to these issues. As described in Section 9.1, IGMP/MLD
proxy mechanisms can address some of these issues. Similarly,
exterior routing protocol messages could be tunneled or conveyed
across an SMF routing domain but doing this robustly in a distributed
wireless environment likely requires additional considerations
outside the scope of this document.
The need for the border router to receive traffic from recognized
multicast sources within the SMF routing domain is important to
achieve interoperability with some existing routing protocols. For
instance, PIM-S requires routers with locally attached multicast
sources to register them to the Rendezvous Point (RP) so that routers
can join the multicast tree. In addition, if those sources are not
advertised to other autonomous systems (ASes) using Multicast Source
Discovery Protocol (MSDP), receivers in those external networks are
not able to join the multicast tree for that source.
9.4. Multiple Border Routers
An SMF routing domain might be deployed with multiple participating
routers having connectivity to external, fixed-infrastructure
networks. Allowing multiple routers to forward multicast traffic to/
from the SMF routing domain can be beneficial since it can increase
reliability and provide better service. For example, if the SMF
routing domain were to fragment with different SMF routers
maintaining connectivity to different border routers, multicast
service could still continue successfully. But, the case of multiple
border routers connecting an SMF routing domain to external networks
presents several challenges for SMF:
1. Handling duplicate unmarked IPv4 or IPv6 (without IPsec
encapsulation or DPD option) packets possibly injected by
multiple border routers.
2. Handling of duplicate traffic injected by multiple border routers
by source-based relay algorithms.
3. Determining which border router(s) will forward outbound
multicast traffic.
4. Additional challenges with interfaces to exterior multicast
routing protocols.
When multiple border routers are present, they may be alternatively
(due to route changes) or simultaneously injecting common traffic
into the SMF routing domain that has not been previously marked for
IPv6 SMF_DPD. Different border routers would not be able to
implicitly synchronize sequencing of injected traffic since they may
not receive exactly the same messages due to packet losses. For IPv6
I-DPD operation, the optional TaggerId field described for the
SMF_DPD option header can be used to mitigate this issue. When
multiple border routers are injecting a flow into an SMF routing
domain, there are two forwarding policies that SMF routers running
I-DPD may implement:
1. Redundantly forward the multicast flows (identified by <srcAddr:
dstAddr>) from each border router, performing DPD processing on a
<TaggerID:dstAddr> or <TaggerID:srcAddr:dstAddr> basis, or
2. Use some basis to select the flow of one tagger (border router)
over the others and forward packets for applicable flows
(identified by <sourceAddress:dstAddr>) only for the selected
TaggerId until timeout or some other criteria to favor another
tagger occurs.
It is RECOMMENDED that the first approach be used in the case of
I-DPD operation. Additional specification may be required to
describe an interoperable forwarding policy based on this second
option. Note that the implementation of the second option requires
that per-flow (i.e., <srcAddr::dstAddr>) state be maintained for the
selected TaggerId.
The deployment of H-DPD operation may alleviate DPD resolution when
ingressing traffic comes from multiple border routers. Non-colliding
hash indexes (those not requiring the H-DPD options header in IPv6)
should be resolved effectively.
10. Security Considerations
Gratuitous use of option headers can cause problems in routers.
Other IP routers external to an SMF routing domain that might receive
forwarded multicast SHOULD ignore SMF-specific IPv6 header options
when encountered. The header option types are encoded appropriately
to allow for this behavior.
This section briefly discusses several SMF denial-of-service (DoS)
attack scenarios and provides some initial recommended mitigation
strategies.
A potential denial-of-service attack against SMF forwarding is
possible when a malicious router has a form of wormhole access to
non-adjacent parts of a network topology. In the wireless ad hoc
case, a directional antenna is one way to provide such a wormhole
physically. If such a router can preview forwarded packets in a non-
adjacent part of the network and forward modified versions to another
part of the network, it can perform the following attack. The
malicious router could reduce the TTL/hop limit or hop limit of the
packet and transmit it to the SMF router causing it to forward the
packet with a limited TTL/hop limit (or even drop it) and make a DPD
entry that could block or limit the subsequent forwarding of later-
arriving valid packets with correct TTL/hop limit values. This would
be a relatively low-cost, high-payoff attack that would be hard to
detect and thus attractive to potential attackers. An approach of
caching TTL/hop limit information with DPD state and taking
appropriate forwarding actions is identified in Section 5 to mitigate
this form of attack.
Sequence-based packet identifiers are predictable and thus provide an
opportunity for a DoS attack against forwarding. Forwarding
protocols that use DPD techniques, such as SMF, may be vulnerable to
DoS attacks based on spoofing packets with apparently valid packet
identifier fields. In wireless environments, where SMF will most
likely be used, the opportunity for such attacks may be more
prevalent than in wired networks. In the case of IPv4 packets,
fragmented IP packets, or packets with IPsec headers applied, the DPD
"identifier portions" of potential future packets that might be
forwarded is highly predictable and easily subject to DoS attacks
against forwarding. A RECOMMENDED technique to counter this concern
is for SMF implementations to generate an "internal" hash value that
is concatenated with the explicit I-DPD packet identifier to form a
unique identifier that is a function of the packet content as well as
the visible identifier. SMF implementations could seed their hash
generation with a random value to make it unlikely that an external
observer could guess how to spoof packets used in a denial-of-service
attack against forwarding. Since the hash computation and state is
kept completely internal to SMF routers, the cryptographic properties
of this hashing would not need to be extensive and thus possibly of
low complexity. Experimental implementations may determine that even
a lightweight hash of only portions of packets may suffice to serve
this purpose.
While H-DPD is not as readily susceptible to this form of DoS attack,
it is possible that a sophisticated adversary could use side
information to construct spoofing packets to mislead forwarders using
a well-known hash algorithm. Thus, similarly, a separate "internal"
hash value could be concatenated with the well-known hash value to
alleviate this security concern.
The support of forwarding IPsec packets without further modification
for both IPv4 and IPv6 is supported by this specification.
Authentication mechanisms to identify the source of IPv6 option
headers should be considered to reduce vulnerability to a variety of
attacks.
Furthermore, when the MANET Neighborhood Discovery Protocol [RFC6130]
is used, the security considerations described in [RFC6130] also
apply.
11. IANA Considerations
This document defines one IPv6 Hop-by-Hop Option, a type for which
has been allocated from the IPv6 "Destination Options and Hop-by-Hop
Options" registry of [RFC2780].
This document creates one registry called "TaggerId Types" for
recording TaggerId types, (TidTy), as a sub-registry in the "IPv6
Parameters" registry.
This document registers one well-known multicast address from each of
the IPv4 and IPv6 multicast address spaces.
This document defines one Message TLV, a type for which has been
allocated from the "Message TLV Types" registry of [RFC5444].
Finally, this document defines one Address Block TLV, a type for
which has been allocated from the "Address Block TLV Types" registry
of [RFC5444].
11.1. IPv6 SMF_DPD Header Extension Option Type
IANA has allocated an IPv6 Option Type from the IPv6 "Destination
Options and Hop-by-Hop Options" registry of [RFC2780], as specified
in Table 9.
+-----------+-------------------------+-------------+---------------+
| Hex Value | Binary Value | Description | Reference |
| | act | chg | rest | | |
+-----------+-------------------------+-------------+---------------+
| 8 | 00 | 0 | 01000 | SMF_DPD | This Document |
+-----------+-------------------------+-------------+---------------+
Table 9: IPv6 Option Type Allocation
11.2. TaggerId Types (TidTy)
A portion of the option data content in the SMF_DPD is the Tagger
Identifier Type (TidTy), which provides a context for the optionally
included TaggerId.
IANA has created a registry for recording TaggerId Types (TidTy),
with initial assignments and allocation policies, as specified in
Table 10.
+------+----------+------------------------------------+------------+
| Type | Mnemonic | Description | Reference |
+------+----------+------------------------------------+------------+
| 0 | NULL | No TaggerId field is present | This |
| | | | document |
| 1 | DEFAULT | A TaggerId of non-specific context | This |
| | | is present | document |
| 2 | IPv4 | A TaggerId representing an IPv4 | This |
| | | address is present | document |
| 3 | IPv6 | A TaggerId representing an IPv6 | This |
| | | address is present | document |
| 4-7 | | Unassigned | |
+------+----------+------------------------------------+------------+
Table 10: TaggerId Types
For allocation of unassigned values 4-7, IETF Review [RFC5226] is
required.
11.3. Well-Known Multicast Address
IANA has allocated an IPv4 multicast address "SL-MANET-ROUTERS"
(224.0.1.186) from the "Internetwork Control Block (224.0.1.0-
224.0.1.255 (224.0.1/24))" sub-registry of the "IPv4 Multicast
Address" registry.
IANA has allocated an IPv6 multicast address "SL-MANET-ROUTERS" from
the "Site-Local Scope Multicast Addresses" sub-sub-registry of the
"Fixed Scope Multicast Addresses" sub-registry of the "INTERNET
PROTOCOL VERSION 6 MULTICAST ADDRESSES" registry.
11.4. SMF TLVs
11.4.1. Expert Review for Created Type Extension Registries
Creation of Address Block TLV Types and Message TLV Types in
registries of [RFC5444], and hence in the HELLO-message-specific
registries of [RFC6130], entails creation of corresponding Type
Extension registries for each such type. For such Type Extension
registries, where an Expert Review is required, the designated expert
SHOULD take the same general recommendations into consideration as
those specified by [RFC5444].
11.4.2. SMF Message TLV Type (SMF_TYPE)
This document defines one Message TLV Type, "SMF_TYPE", which has
been allocated from the "HELLO Message-Type-specific Message TLV
Types" registry, defined in [RFC6130].
This created a new Type Extension registry, with initial assignments
as specified in Table 11.
+----------+------+-----------+--------------------+----------------+
| Name | Type | Type | Description | Allocation |
| | | Extension | | Policy |
+----------+------+-----------+--------------------+----------------+
| SMF_TYPE | 128 | 0-255 | Specifies relay | Section 11.4.4 |
| | | | algorithm | |
| | | | supported by the | |
| | | | SMF router, | |
| | | | originating the | |
| | | | HELLO message, | |
| | | | according to | |
| | | | Section 11.4.4. | |
+----------+------+-----------+--------------------+----------------+
Table 11: SMF_TYPE Message TLV Type Extension Registry
11.4.3. SMF Address Block TLV Type (SMF_NBR_TYPE)
This document defines one Address Block TLV Type, "SMF_NBR_TYPE",
which has been allocated from the "HELLO Message-Type-specific
Address Block TLV Types" registry, defined in [RFC6130].
This has created a new Type Extension registry, with initial
assignments as specified in Table 12.
+--------------+--------+-----------+-----------------+-------------+
| Name | Type | Type | Description | Allocation |
| | | Extension | | Policy |
+--------------+--------+-----------+-----------------+-------------+
| SMF_NBR_TYPE | 128 | 0-255 | Specifies relay | Section |
| | | | algorithm | 11.4.4 |
| | | | supported by | |
| | | | the SMF router | |
| | | | corresponding | |
| | | | to the | |
| | | | advertised | |
| | | | address, | |
| | | | according to | |
| | | | Section 11.4.4. | |
+--------------+--------+-----------+-----------------+-------------+
Table 12: SMF_NBR_TYPE Address Block TLV Type Extension Registry
11.4.4. SMF Relay Algorithm ID Registry
Types for the Type Extension Registries for the SMF_TYPE Message TLV
and the SMF_NBR_TYPE Address Block TLV are unified in this single SMF
Relay Algorithm ID Registry, defined in this section.
IANA has created a registry for recording Relay Algorithm
Identifiers, with initial assignments and allocation policies as
specified in Table 13.
+---------+---------+-------------+-------------------+
| Value | Name | Description | Allocation Policy |
+---------+---------+-------------+-------------------+
| 0 | CF | Section 4 | |
| 1 | S-MPR | Appendix B | |
| 2 | E-CDS | Appendix A | |
| 3 | MPR-CDS | Appendix C | |
| 4-127 | | Unassigned | Expert Review |
| 128-255 | | Unassigned | Experimental Use |
+---------+---------+-------------+-------------------+
Table 13: Relay Set Algorithm Type Values
A specification requesting an allocation from the 4-127 range from
the SMF Relay Algorithm ID Registry MUST specify the interpretation
of the <value> field (if any).
12. Acknowledgments
Many of the concepts and mechanisms used and adopted by SMF resulted
over several years of discussion and related work within the MANET
working group since the late 1990s. There are obviously many
contributors to past discussions and related draft documents within
the working group that have influenced the development of SMF
concepts, and they deserve acknowledgment. In particular, this
document is largely a direct product of the earlier SMF design team
within the IETF MANET working group and borrows text and
implementation ideas from the related individuals and activities.
Some of the direct contributors who have been involved in design,
content editing, prototype implementation, major commenting, and core
discussions are listed below in alphabetical order. We appreciate
all the input and feedback from the many community members and early
implementation users we have heard from that are not on this list as
well.
Brian Adamson
Teco Boot
Ian Chakeres
Thomas Clausen
Justin Dean
Brian Haberman
Ulrich Herberg
Charles Perkins
Pedro Ruiz
Fred Templin
Maoyu Wang
13. References
13.1. Normative References
[MPR-CDS] Adjih, C., Jacquet, P., and L. Viennot, "Computing
Connected Dominating Sets with Multipoint Relays", Ad Hoc
and Sensor Wireless Networks, January 2005.
[RFC0791] Postel, J., "Internet Protocol", STD 5, RFC 791,
September 1981.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC2460] Deering, S. and R. Hinden, "Internet Protocol, Version 6
(IPv6) Specification", RFC 2460, December 1998.
[RFC2644] Senie, D., "Changing the Default for Directed Broadcasts
in Routers", BCP 34, RFC 2644, August 1999.
[RFC2780] Bradner, S. and V. Paxson, "IANA Allocation Guidelines For
Values In the Internet Protocol and Related Headers",
BCP 37, RFC 2780, March 2000.
[RFC3174] Eastlake, D. and P. Jones, "US Secure Hash Algorithm 1
(SHA1)", RFC 3174, September 2001.
[RFC3626] Clausen, T. and P. Jacquet, "Optimized Link State Routing
Protocol (OLSR)", RFC 3626, October 2003.
[RFC4291] Hinden, R. and S. Deering, "IP Version 6 Addressing
Architecture", RFC 4291, February 2006.
[RFC4302] Kent, S., "IP Authentication Header", RFC 4302,
December 2005.
[RFC5226] Narten, T. and H. Alvestrand, "Guidelines for Writing an
IANA Considerations Section in RFCs", BCP 26, RFC 5226,
May 2008.
[RFC5444] Clausen, T., Dearlove, C., Dean, J., and C. Adjih,
"Generalized Mobile Ad Hoc Network (MANET) Packet/Message
Format", RFC 5444, February 2009.
[RFC5614] Ogier, R. and P. Spagnolo, "Mobile Ad Hoc Network (MANET)
Extension of OSPF Using Connected Dominating Set (CDS)
Flooding", RFC 5614, August 2009.
[RFC5771] Cotton, M., Vegoda, L., and D. Meyer, "IANA Guidelines for
IPv4 Multicast Address Assignments", BCP 51, RFC 5771,
March 2010.
[RFC6130] Clausen, T., Dearlove, C., and J. Dean, "Mobile Ad Hoc
Network (MANET) Neighborhood Discovery Protocol (NHDP)",
RFC 6130, April 2011.
13.2. Informative References
[CDHM07] Chakeres, I., Danilov, C., Henderson, T., and J. Macker,
"Connecting MANET Multicast", IEEE MILCOM
2007 Proceedings, 2007.
[DHG09] Danilov, C., Henderson, T., Goff, T., Kim, J., Macker, J.,
Weston, J., Neogi, N., Ortiz, A., and D. Uhlig,
"Experiment and field demonstration of a 802.11-based
ground-UAV mobile ad-hoc network", Proceedings of the 28th
IEEE conference on Military Communications, 2009.
[DHS08] Danilov, C., Henderson, T., Spagnolo, T., Goff, T., and J.
Kim, "MANET Multicast with Multiple Gateways", IEEE MILCOM
2008 Proceedings, 2008.
[GM99] Garcia-Luna-Aceves, JJ. and E. Madruga, "The Core-Assisted
Mesh Protocol", Selected Areas in Communications, IEEE
Journal, Volume 17, Issue 8, August 1999.
[IPV4-ID-UPDATE]
Touch, J., "Updated Specification of the IPv4 ID Field",
Work in Progress, September 2011.
[JLMV02] Jacquet, P., Laouiti, V., Minet, P., and L. Viennot,
"Performance of Multipoint Relaying in Ad Hoc Mobile
Routing Protocols", Networking , 2002.
[MDC04] Macker, J., Dean, J., and W. Chao, "Simplified Multicast
Forwarding in Mobile Ad hoc Networks", IEEE MILCOM 2004
Proceedings, 2004.
[MDDA07] Macker, J., Downard, I., Dean, J., and R. Adamson,
"Evaluation of Distributed Cover Set Algorithms in Mobile
Ad hoc Network for Simplified Multicast Forwarding", ACM
SIGMOBILE Mobile Computing and Communications
Review, Volume 11, Issue 3, July 2007.
[MGL04] Mohapatra, P., Gui, C., and J. Li, "Group Communications
in Mobile Ad hoc Networks", IEEE Computer, Vol. 37, No. 2,
February 2004.
[NTSC99] Ni, S., Tseng, Y., Chen, Y., and J. Sheu, "The Broadcast
Storm Problem in a Mobile Ad Hoc Network", Proceedings of
ACM Mobicom 99, 1999.
[RFC2501] Corson, M. and J. Macker, "Mobile Ad hoc Networking
(MANET): Routing Protocol Performance Issues and
Evaluation Considerations", RFC 2501, January 1999.
[RFC3684] Ogier, R., Templin, F., and M. Lewis, "Topology
Dissemination Based on Reverse-Path Forwarding (TBRPF)",
RFC 3684, February 2004.
[RFC3973] Adams, A., Nicholas, J., and W. Siadak, "Protocol
Independent Multicast - Dense Mode (PIM-DM): Protocol
Specification (Revised)", RFC 3973, January 2005.
[RFC4601] Fenner, B., Handley, M., Holbrook, H., and I. Kouvelas,
"Protocol Independent Multicast - Sparse Mode (PIM-SM):
Protocol Specification (Revised)", RFC 4601, August 2006.
Appendix A. Essential Connecting Dominating Set (E-CDS) Algorithm
The "Essential Connected Dominating Set" (E-CDS) algorithm [RFC5614]
forms a single CDS mesh for the SMF operating region. It allows
routers to use 2-hop neighborhood topology information to dynamically
perform relay self-election to form a CDS. Its packet-forwarding
rules are not dependent upon previous hop knowledge. Additionally,
E-CDS SMF forwarders can be easily mixed without problems with CF SMF
forwarders, even those not participating in NHDP. Another benefit is
that packets opportunistically received from non-symmetric neighbors
may be forwarded without compromising flooding efficiency or
correctness. Furthermore, multicast sources not participating in
NHDP may freely inject their traffic, and any neighboring E-CDS
relays will properly forward the traffic. The E-CDS-based relay set
selection algorithm is based upon [RFC5614]. E-CDS was originally
discussed in the context of forming partial adjacencies and efficient
flooding for MANET OSPF extensions work, and the core algorithm is
applied here for SMF.
It is RECOMMENDED that the SMF_TYPE:E-CDS Message TLV be included in
NHDP_HELLO messages that are generated by routers conducting E-CDS
SMF operation. It is also RECOMMENDED that the SMF_NBR_TYPE:E-CDS
Address Block TLV be used to advertise neighbor routers that are also
conducting E-CDS SMF operation.
A.1. E-CDS Relay Set Selection Overview
The E-CDS relay set selection requires 2-hop neighborhood information
collected through NHDP or another process. Relay routers, in E-CDS
SMF selection, are "self-elected" using a Router Identifier (Router
ID) and an optional nodal metric, referred to here as Router Priority
for all 1-hop and 2-hop neighbors. To ensure proper relay set self-
election, the Router ID and Router Priority MUST be consistent among
participating routers. It is RECOMMENDED that NHDP be used to share
Router ID and Router Priority through the use of SMF_TYPE:E-CDS TLVs
as described in this appendix. The Router ID is a logical
identification that MUST be consistent across interoperating SMF
neighborhoods, and it is RECOMMENDED to be chosen as the numerically
largest address contained in a router's "Neighbor Address List" as
defined in NHDP. The E-CDS self-election process can be summarized
as follows:
1. If an SMF router has a higher ordinal (Router Priority, Router
ID) than all of its symmetric neighbors, it elects itself to act
as a forwarder for all received multicast packets.
2. Else, if there does not exist a path from the neighbor with
largest (Router Priority, Router ID) to any other neighbor, via
neighbors with larger values of (Router Priority, Router ID),
then it elects itself to the relay set.
The basic form of E-CDS described and applied within this
specification does not provide for redundant relay set selection
(e.g., bi-connected), but such capability is supported by the basic
E-CDS design.
A.2. E-CDS Forwarding Rules
With E-CDS, any SMF router that has selected itself as a relay
performs DPD and forwards all non-duplicative multicast traffic
allowed by the present forwarding policy. Packet previous-hop
knowledge is not needed for forwarding decisions when using E-CDS.
1. Upon packet reception, DPD is performed. Note E-CDS requires a
single duplicate table for the set of interfaces associated with
the relay set selection.
2. If the packet is a duplicate, no further action is taken.
3. If the packet is non-duplicative:
A. A DPD entry is made for the packet identifier.
B. The packet is forwarded out to all interfaces associated with
the relay set selection.
As previously mentioned, even packets sourced (or relayed) by routers
not participating in NHDP and/or the E-CDS relay set selection may be
forwarded by E-CDS forwarders without problem. A particular
deployment MAY choose to not forward packets from previous hop
routers that have been not explicitly identified via NHDP or other
means as operating as part of a different relay set algorithm (e.g.,
S-MPR) to allow coexistent deployments to operate correctly. Also,
E-CDS relay set selection may be configured to be influenced by
statically configured CF relays that are identified via NHDP or other
means.
A.3. E-CDS Neighborhood Discovery Requirements
It is possible to perform E-CDS relay set selection without
modification of NHDP, basing the self-election process exclusively on
the "Neighbor Address List" of participating SMF routers, for
example, by setting the Router Priority to a default value and
selecting the Router ID as the numerically largest address contained
in the "Neighbor Address List". However, steps MUST be taken to
ensure that all NHDP-enabled routers not using SMF_TYPE:E-CDS full
type Message TLVs are, in fact, running SMF E-CDS with the same
methods for selecting Router Priority and Router ID; otherwise,
incorrect forwarding may occur. Note that SMF routers with higher
Router Priority values will be favored as relays over routers with
lower Router Priority. Thus, preferred relays MAY be
administratively configured to be selected when possible.
Additionally, other metrics (e.g., nodal degree, energy capacity,
etc.) may also be taken into account in constructing a Router
Priority value. When using Router Priority with multiple interfaces,
all interfaces on a router MUST use and advertise a common Router
Priority value. A router's Router Priority value may be
administratively or algorithmically selected. The method of
selection does not need to be the same among different routers.
E-CDS relay set selection may be configured to be influenced by
statically configured CF relays that are identified via NHDP or other
means. Nodes advertising CF through NHDP may be considered E-CDS SMF
routers with maximal Router Priority.
To share a router's Router Priority with its 1-hop neighbors, the
SMF_TYPE:E-CDS Message TLV's <value> field is defined as shown in
Table 14.
+----------------+---------+-----------------+
| Length (bytes) | Value | Router Priority |
+----------------+---------+-----------------+
| 0 | N/A | 64 |
| 1 | <value> | 0-127 |
+----------------+---------+-----------------+
Table 14: E-CDS Message TLV Values
Where <value> is a one-octet-long bit field that is defined as:
bit 0: the leftmost bit is reserved and SHOULD be set to 0.
bits 1-7: contain the unsigned Router Priority value, 0-127, which is
associated with the "Neighbor Address List".
Combinations of value field lengths and values other than specified
here are NOT permitted and SHOULD be ignored. Figure 6 shows an
example SMF_TYPE:E-CDS Message 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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... | SMF_TYPE |1|0|0|1|0|0| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| E-CDS |0|0|0|0|0|0|0|1|R| priority | ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 6: E-CDS Message TLV Example
To convey Router Priority values among 2-hop neighborhoods, the
SMF_NBR_TYPE:E-CDS Address Block TLV's <value> field is used. Multi-
index and multivalue TLV layouts as defined in [RFC5444] are
supported. SMF_NBR_TYPE:E-CDS value fields are defined thus:
+---------------+--------+----------+-------------------------------+
| Length(bytes) | # Addr | Value | Router Priority |
+---------------+--------+----------+-------------------------------+
| 0 | Any | N/A | 64 |
| 1 | Any | <value> | <value> is for all addresses |
| N | N | <value>* | Each address gets its own |
| | | | <value> |
+---------------+--------+----------+-------------------------------+
Table 15: E-CDS Address Block TLV Values
Where <value> is a one-byte bit field that is defined as:
bit 0: the leftmost bit is reserved and SHOULD be set to 0.
bits 1-7: contain the unsigned Router Priority value, 0-127, which is
associated with the appropriate address(es).
Combinations of value field lengths and # of addresses other than
specified here are NOT permitted and SHOULD be ignored. A default
technique of using nodal degree (i.e., count of 1-hop neighbors) is
RECOMMENDED for the value field of these TLV types. Below are two
example SMF_NBR_TYPE:E-CDS Address Block TLVs.
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... | SMF_NBR_TYPE |1|0|0|1|0|0| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| E-CDS |0|0|0|0|0|0|0|1|R| priority | ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 7: E-CDS Address Block TLV Example 1
The single value example TLV, depicted in Figure 7, specifies that
all address(es) contained in the address block are running SMF using
the E-CDS algorithm and all address(es) share the value field and
therefore the same Router Priority.
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... | SMF_NBR_TYPE |1|0|1|1|0|1| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| E-CDS | index-start | index-end | length |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|R| priority0 |R| priority1 | ... |R| priorityN |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 8: E-CDS Address Block TLV Example 2
The example multivalued TLV, depicted in Figure 8, specifies that
address(es) contained in the address block from index-start to index-
end inclusive are running SMF using the E-CDS algorithm. Each
address is associated with its own value byte and therefore its own
Router Priority.
A.4. E-CDS Selection Algorithm
This section describes an algorithm for E-CDS relay selection (self-
election). The algorithm described uses 2-hop information. Note
that it is possible to extend this algorithm to use k-hop information
with added computational complexity and mechanisms for sharing k-hop
topology information that are not described in this document or
within the NHDP specification. It should also be noted that this
algorithm does not impose the hop limit bound described in [RFC5614]
when performing the path search that is used for relay selection.
However, the algorithm below could be easily augmented to accommodate
this additional criterion. It is not expected that the hop limit
bound will provide significant benefit to the algorithm defined in
this appendix.
The tuple of Router Priority and Router ID is used in E-CDS relay set
selection. Precedence is given to the Router Priority portion, and
the Router ID value is used as a tiebreaker. The evaluation of this
tuple is referred to as "RtrPri(n)" in the description below where
"n" references a specific router. Note that it is possible that the
Router Priority portion may be optional and the evaluation of
"RtrPri()" be solely based upon the unique Router ID. Since there
MUST NOT be any duplicate Router ID values among SMF routers, a
comparison of "RtrPri(n)" between any two routers will always be an
inequality. The use of nodal degree for calculating Router Priority
is RECOMMENDED as default, and the largest IP address in the
"Neighbor Address List" as advertised by NHDP MUST be used as the
Router ID. NHDP provides all interface addresses throughout the
2-hop neighborhood through HELLO messages, so explicitly conveying a
Router ID is not necessary. The following steps describe a basic
algorithm for conducting E-CDS relay selection for a router "n0":
1. Initialize the set "N1" with tuples ("Router Priority", "Router
ID", "Neighbor Address List") for each 1-hop neighbor of "n0".
2. If "N1" has less than 2 tuples, then "n0" does not elect itself
as a relay, and no further steps are taken.
3. Initialize the set "N2" with tuples ("Router Priority", "Router
ID", "2-hop address") for each "2-hop address" of "n0", where
"2-hop address" is defined in NHDP.
4. If "RtrPri(n0)" is greater than that of all tuples in
"N1", then "n0" selects itself as a relay, and no
further steps are taken.
EID 3640 (Verified) is as follows:Section: A.4
Original Text:
4. If "RtrPri(n0)" is greater than that of all tuples in the union
of "N1" and "N2", then "n0" selects itself as a relay, and no
further steps are taken.
Corrected Text:
4. If "RtrPri(n0)" is greater than that of all tuples in
"N1", then "n0" selects itself as a relay, and no
further steps are taken.
Notes:
The initial verbal description of the E-CDS algorithm in first paragraph A.1 pg 40 is correct..as follows
1. If an SMF router has a higher ordinal (Router Priority, Router ID) than all of its symmetric neighbors, it elects itself to act as a forwarder for all received multicast packets.
But LATER in A.4 pseudocode Step 4 contains a bug. N2 (2 hop neighbors are included in this step). This pseudocode bug can cause incorrect behavior to occur.
5. Initialize all tuples in the union of "N1" and "N2" as
"unvisited".
6. Find the tuple "n1_Max" that has the largest "RtrPri()" of all
tuples in "N1".
7. Initialize queue "Q" to contain "n1_Max", marking "n1_Max" as
"visited".
8. While router queue "Q" is not empty, remove router "x" from the
head of "Q", and for each 1-hop neighbor "n" of router "x"
(excluding "n0") that is not marked "visited".
A. Mark router "n" as "visited".
B. If "RtrPri(n)" is greater than "RtrPri(n0)", append "n" to
"Q".
9. If any tuple in "N1" remains "unvisited", then "n0" selects
itself as a relay. Otherwise, "n0" does not act as a relay.
Note these steps are re-evaluated upon neighborhood status changes.
Steps 5 through 8 of this procedure describe an approach to a path
search. The purpose of this path search is to determine if paths
exist from the 1-hop neighbor with maximum "RtrPri()" to all other
1-hop neighbors without traversing an intermediate router with a
"RtrPri()" value less than "RtrPri(n0)". These steps comprise a
breadth-first traversal that evaluates only paths that meet that
criteria. If all 1-hop neighbors of "n0" are "visited" during this
traversal, then the path search has succeeded, and router "n0" does
not need to provide relay. It can be assumed that other routers will
provide relay operation to ensure SMF connectivity.
It is possible to extend this algorithm to consider neighboring SMF
routers that are known to be statically configured for CF (always
relaying). The modification to the above algorithm is to process
such routers as having a maximum possible Router Priority value. It
is expected that routers configured for CF and participating in NHDP
would indicate this with use of the SMF_TYPE:CF and SMF_NBR_TYPE:CF
TLV types in their NHDP_HELLO message and address blocks,
respectively.
Appendix B. Source-Based Multipoint Relay (S-MPR) Algorithm
The source-based multipoint relay (S-MPR) set selection algorithm
enables individual routers, using 2-hop topology information, to
select relays from their set of neighboring routers. Relays are
selected so that forwarding to the router's complete 2-hop neighbor
set is covered. This distributed relay set selection technique has been shown to
approximate a minimum connected dominating set (MCDS) in [JLMV02]. Individual routers must collect 2-hop neighborhood
EID 4098 (Verified) is as follows:Section: Appendix B
Original Text:
This distributed relay set selection technique has been shown to
approximate a minimal connected dominating set (MCDS) in [JLMV02].
Corrected Text:
This distributed relay set selection technique has been shown to
approximate a minimum connected dominating set (MCDS) in [JLMV02].
Notes:
Minimum connected dominating set [1] is the established terminology and minimal was an editorial error.
[1] Sampathkumar, E.; Walikar, HB (1979), "The connected domination number of a graph", J. Math. Phys. Sci 13 (6): 607–613.
information from neighbors, determine an appropriate current relay
set, and inform selected neighbors of their relay status. Note that
since each router picks its neighboring relays independently, S-MPR
forwarders depend upon previous hop information (e.g., source MAC
address) to operate correctly. The Optimized Link State Routing
(OLSR) protocol has used this algorithm and protocol for relay of
link state updates and other control information [RFC3626], and it
has been demonstrated operationally in dynamic network environments.
It is RECOMMENDED that the SMF_TYPE:S-MPR Message TLV be included in
NHDP_HELLO messages that are generated by routers conducting S-MPR
SMF operation. It is also RECOMMENDED that the SMF_NBR_TYPE:S-MPR
Address Block TLV be used to specify which neighbor routers are
conducting S-MPR SMF operation.
B.1. S-MPR Relay Set Selection Overview
The S-MPR algorithm uses bi-directional 1-hop and 2-hop neighborhood
information collected via NHDP to select, from a router's 1-hop
neighbors, a set of relays that will cover the router's entire 2-hop
neighbor set upon forwarding. The algorithm described uses a
"greedy" heuristic of first picking the 1-hop neighbor who will cover
the most 2-hop neighbors. Then, excluding those 2-hop neighbors that
have been covered, additional relays from its 1-hop neighbor set are
iteratively selected until the entire 2-hop neighborhood is covered.
Note that 1-hop neighbors also identified as 2-hop neighbors are
considered as 1-hop neighbors only.
NHDP HELLO messages supporting S-MPR forwarding operation SHOULD use
the TLVs defined in Section 8.1 using the S-MPR extended type. The
value field of an Address Block TLV that has a full type value of
SMF_NBR_TYPE:S-MPR is defined in Table 17 such that signaling of MPR
selections to 1-hop neighbors is possible. The value field of a
message block TLV that has a full type value of SMF_TYPE:S-MPR is
defined in Table 16 such that signaling of Router Priority (described
as "WILLINGNESS" in [RFC3626]) to 1-hop neighbors is possible. It is
important to note that S-MPR forwarding is dependent upon the
previous hop of an incoming packet. An S-MPR router MUST forward
packets only for neighbors that have explicitly selected it as a
multipoint relay (i.e., its "selectors"). There are also some
additional requirements for duplicate packet detection to support
S-MPR SMF operation that are described below.
For multiple interface operation, MPR selection SHOULD be conducted
on a per-interface basis. However, it is possible to economize MPR
selection among multiple interfaces by selecting common MPRs to the
extent possible.
B.2. S-MPR Forwarding Rules
An S-MPR SMF router MUST only forward packets for neighbors that have
explicitly selected it as an MPR. The source-based forwarding
technique also stipulates some additional duplicate packet detection
operations. For multiple network interfaces, independent DPD state
MUST be maintained for each separate interface. The following
provides the procedure for S-MPR packet forwarding given the arrival
of a packet on a given interface, denoted <srcIface>. There are
three possible actions, depending upon the previous-hop transmitter:
1. If the previous-hop transmitter has selected the current router
as an MPR,
A. The packet identifier is checked against the DPD state for
each possible outbound interface, including the <srcIface>.
B. If the packet is not a duplicate for an outbound interface,
the packet is forwarded on that interface and a DPD entry is
made for the given packet identifier for the interface.
C. If the packet is a duplicate, no action is taken for that
interface.
2. Else, if the previous-hop transmitter is a 1-hop symmetric
neighbor, a DPD entry is added for that packet for the
<srcIface>, but the packet is not forwarded.
3. Otherwise, no action is taken.
Action number two in the list above is non-intuitive but important to
ensure correctness of S-MPR SMF operation. The selection of source-
based relays does not result in a common set among neighboring
routers, so relays MUST mark, in their DPD state, packets received
from non-selector, symmetric, 1-hop neighbors (for a given interface)
and not forward subsequent duplicates of that packet if received on
that interface. Deviation here can result in unnecessary, repeated
packet forwarding throughout the network or incomplete flooding.
Nodes not participating in neighborhood discovery and relay set
selection will not be able to source multicast packets into the area
and have SMF forward them, unlike E-CDS or MPR-CDS where forwarding
may occur dependent on topology. Correct S-MPR relay behavior will
occur with the introduction of repeaters (non-NHDP/SMF participants
that relay multicast packets using duplicate detection and CF), but
the repeaters will not efficiently contribute to S-MPR forwarding as
these routers will not be identified as neighbors (symmetric or
otherwise) in the S-MPR forwarding process. NHDP/SMF participants
MUST NOT forward packets that are not selected by the algorithm, as
this can disrupt network-wide S-MPR flooding, resulting in incomplete
or inefficient flooding. The result is that non-S-MPR SMF routers
will be unable to source multicast packets and have them forwarded by
other S-MPR SMF routers.
B.3. S-MPR Neighborhood Discovery Requirements
Nodes may optionally signal a Router Priority value to their 1-hop
neighbors by using the SMF_TYPE:S-MPR message block TLV value field.
If the value field is omitted, a default Router Priority value of 64
is to be assumed. This is summarized here:
+---------------+---------+-----------------+
| Length(bytes) | Value | Router Priority |
+---------------+---------+-----------------+
| 0 | N/A | 64 |
| 1 | <value> | 0-127 |
+---------------+---------+-----------------+
Table 16: S-MPR Message TLV Values
Where <value> is a one-octet-long bit field defined as:
bit 0: the leftmost bit is reserved and SHOULD be set to 0.
bits 1-7: contain the Router Priority value, 0-127, which is
associated with the "Neighbor Address List".
Router Priority values for S-MPR are interpreted in the same fashion
as "WILLINGNESS" ([RFC3626]), with the value 0 indicating a router
will NEVER forward and value 127 indicating a router will ALWAYS
forward. Values 1-126 indicate how likely a S-MPR SMF router will be
selected as an MPR by a neighboring SMF router, with higher values
increasing the likelihood. Combinations of value field lengths and
values other than those specified here are NOT permitted and SHOULD
be ignored. Below is an example SMF_TYPE:S-MPR Message 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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... | SMF_TYPE |1|0|0|1|0|0| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| S-MPR |0|0|0|0|0|0|0|1|R| priority | ... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 9: S-MPR Message TLV Example
S-MPR election operation requires 2-hop neighbor knowledge as
provided by NHDP [RFC6130] or from external sources. MPRs are
dynamically selected by each router, and selections MUST be
advertised and dynamically updated within NHDP or an equivalent
protocol or mechanism. For NHDP use, the SMF_NBR_TYPE:S-MPR Address
Block TLV value field is defined as such:
+---------------+--------+----------+-------------------------------+
| Length(bytes) | # Addr | Value | Meaning |
+---------------+--------+----------+-------------------------------+
| 0 | Any | N/A | NOT MPRs |
| 1 | Any | <value> | <value> is for all addresses |
| N | N | <value>* | Each address gets its own |
| | | | <value> |
+---------------+--------+----------+-------------------------------+
Table 17: S-MPR Address Block TLV Values
Where <value>, if present, is a one-octet bit field defined as:
bit 0: The leftmost bit is the M bit that, when set, indicates MPR
selection of the relevant interface, represented by the associated
address(es), by the originator router of the NHDP HELLO message.
When unset, it indicates the originator router of the NHDP HELLO
message has not selected the relevant interfaces, represented by the
associated address(es), as its MPR.
bits 1-7: These bits are reserved and SHOULD be set to 0.
Combinations of value field lengths and number of addresses other
than specified here are NOT permitted and SHOULD be ignored. All
bits, excepting the leftmost bit, are RESERVED and SHOULD be set to
0.
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
... | SMF_NBR_TYPE |1|1|0|1|0|0| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| S-MPR | start-index |0|0|0|0|0|0|0|1|M| reserved |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 10: S-MPR Address Block TLV Example
The single index TLV example, depicted in Figure 10, indicates that
the address specified by the <start-index> field is running SMF using
S-MPR and has been selected by the originator of the NHDP HELLO
message as an MPR forwarder if the M bit is set. Multivalued TLVs
may also be used to specify MPR selection status of multiple
addresses using only one TLV. See Figure 8 for a similar example on
how this may be done.
B.4. S-MPR Selection Algorithm
This section describes a basic algorithm for the S-MPR selection
process. Note that the selection is with respect to a specific
interface of the router performing selection, and other router
interfaces referenced are reachable from this reference router
interface. This is consistent with the S-MPR forwarding rules
described above. When multiple interfaces per router are used, it is
possible to enhance the overall selection process across multiple
interfaces such that common routers are selected as MPRs for each
interface to avoid unnecessary inefficiencies in flooding. The
following steps describe a basic algorithm for conducting S-MPR
selection for a router interface "n0":
1. Initialize the set "MPR" to empty.
2. Initialize the set "N1" to include all 1-hop neighbors of "n0".
3. Initialize the set "N2" to include all 2-hop neighbors, excluding
"n0" and any routers in "N1". Nodes that are only reachable via
"N1" routers with router priority values of NEVER are also
excluded.
4. For each interface "y" in "N1", initialize a set "N2(y)" to
include any interfaces in "N2" that are 1-hop neighbors of "y".
5. For each interface "x" in "N1" with a router priority value of
"ALWAYS" (or using the CF relay algorithm), select "x" as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
6. For each interface "z" in "N2", initialize the set "N1(z)" to
include any interfaces in "N1" that are 1-hop neighbors of "z".
7. For each interface "w" in "N2" where "N1(w)" has only one member, "x",
select "x" as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
8. While "N2" is not empty, select the interface "x" in "N1" with
the highest router priority [break ties in favor of the node with the
largest number of members in "N_2(x)"] as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
EID 3499 (Verified) is as follows:Section: B.4.
Original Text:
1. Initialize the set "MPR" to empty.
2. Initialize the set "N1" to include all 1-hop neighbors of "n0".
3. Initialize the set "N2" to include all 2-hop neighbors, excluding
"n0" and any routers in "N1". Nodes that are only reachable via
"N1" routers with router priority values of NEVER are also
excluded.
4. For each interface "y" in "N1", initialize a set "N2(y)" to
include any interfaces in "N2" that are 1-hop neighbors of "y".
5. For each interface "x" in "N1" with a router priority value of
"ALWAYS" (or using the CF relay algorithm), select "x" as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
6. For each interface "z" in "N2", initialize the set "N1(z)" to
include any interfaces in "N1" that are 1-hop neighbors of "z".
7. For each interface "x" in "N2" where "N1(x)" has only one member,
select "x" as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2" and
delete "N1(z)".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
8. While "N2" is not empty, select the interface "x" in "N1" with
the largest router priority that has the number of members in
"N_2(x)" as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
Corrected Text:
1. Initialize the set "MPR" to empty.
2. Initialize the set "N1" to include all 1-hop neighbors of "n0".
3. Initialize the set "N2" to include all 2-hop neighbors, excluding
"n0" and any routers in "N1". Nodes that are only reachable via
"N1" routers with router priority values of NEVER are also
excluded.
4. For each interface "y" in "N1", initialize a set "N2(y)" to
include any interfaces in "N2" that are 1-hop neighbors of "y".
5. For each interface "x" in "N1" with a router priority value of
"ALWAYS" (or using the CF relay algorithm), select "x" as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
6. For each interface "z" in "N2", initialize the set "N1(z)" to
include any interfaces in "N1" that are 1-hop neighbors of "z".
7. For each interface "w" in "N2" where "N1(w)" has only one member, "x",
select "x" as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
8. While "N2" is not empty, select the interface "x" in "N1" with
the highest router priority [break ties in favor of the node with the
largest number of members in "N_2(x)"] as an MPR:
A. Add "x" to the set "MPR" and remove "x" from "N1".
B. For each interface "z" in "N2(x)", remove "z" from "N2".
C. For each interface "y" in "N1", remove any interfaces in
"N2(x)" from "N2(y)".
Notes:
There are three changes:
On line 7, the first and second occurrences of x are replaced by w, and then x is given as the name of the sole member of N1(w).
On line 7B, the phrase 'delete "N1(z)" is dropped to be consistent with the rest of the algorithm.
On line 8 some rewording is done for clarification.
This errata prepared in consultation with Justin Dean and Gus Macker.
After the set of routers "MPR" is selected, router "n_0" must signal
its selections to its neighbors. With NHDP, this is done by using
the MPR Address Block TLV to mark selected neighbor addresses in
NHDP_HELLO messages. Neighbors MUST record their MPR selection
status and the previous hop address (e.g., link or MAC layer) of the
selector. Note these steps are re-evaluated upon neighborhood status
changes.
Appendix C. Multipoint Relay Connected Dominating Set (MPR-CDS)
Algorithm
The MPR-CDS algorithm is an extension to the basic S-MPR election
algorithm that results in a shared (non-source-specific) SMF CDS.
Thus, its forwarding rules are not dependent upon previous hop
information, similar to E-CDS. An overview of the MPR-CDS selection
algorithm is provided in [MPR-CDS].
It is RECOMMENDED that the SMF_TYPE Message TLV be included in
NHDP_HELLO messages that are generated by routers conducting MPR-CDS
SMF operation.
C.1. MPR-CDS Relay Set Selection Overview
The MPR-CDS relay set selection process is based upon the MPR
selection process of the S-MPR algorithm with the added refinement of
a distributed technique for subsequently down-selecting to a common
reduced, shared relay set. A router ordering (or "prioritization")
metric is used as part of this down-selection process; like the E-CDS
algorithm, this metric can be based upon router address(es) or some
other unique router identifier (e.g., Router ID based on largest
address contained within the "Neighbor Address List") as well as an
additional Router Priority measure, if desired. The process for MPR-
CDS relay selection is as follows:
1. First, MPR selection per the S-MPR algorithm is conducted, with
selectors informing their MPRs (via NHDP) of their selection.
2. Then, the following rules are used on a distributed basis by
selected routers to possibly deselect themselves and thus jointly
establish a common set of shared SMF relays:
A. If a selected router has a larger "RtrPri()" than all of its
1-hop symmetric neighbors, then it acts as a relay for all
multicast traffic, regardless of the previous hop.
B. Else, if the 1-hop symmetric neighbor with the largest
"RtrPri()" value has selected the router, then it also acts
as a relay for all multicast traffic, regardless of the
previous hop.
C. Otherwise, it deselects itself as a relay and does not
forward any traffic unless changes occur that require re-
evaluation of the above steps.
This technique shares many of the desirable properties of the E-CDS
technique with regards to compatibility with multicast sources not
participating in NHDP and the opportunity for statically configured
CF routers to be present, regardless of their participation in NHDP.
C.2. MPR-CDS Forwarding Rules
The forwarding rules for MPR-CDS are similar to those for E-CDS. Any
SMF router that has selected itself as a relay performs DPD and
forwards all non-duplicative multicast traffic allowed by the present
forwarding policy. Packet previous hop knowledge is not needed for
forwarding decisions when using MPR-CDS.
1. Upon packet reception, DPD is performed. Note that MPR-CDS
requires one duplicate table for the set of interfaces associated
with the relay set selection.
2. If the packet is a duplicate, no further action is taken.
3. If the packet is non-duplicative:
A. A DPD entry is added for the packet identifier
B. The packet is forwarded out to all interfaces associated with
the relay set selection.
As previously mentioned, even packets sourced (or relayed) by routers
not participating in NHDP and/or the MPR-CDS relay set selection may
be forwarded by MPR-CDS forwarders without problem. A particular
deployment MAY choose to not forward packets from sources or relays
that have been explicitly identified via NHDP or other means as
operating as part of a different relay set algorithm (e.g., S-MPR) to
allow coexistent deployments to operate correctly.
C.3. MPR-CDS Neighborhood Discovery Requirements
The neighborhood discovery requirements for MPR-CDS have commonality
with both the S-MPR and E-CDS algorithms. MPR-CDS selection
operation requires 2-hop neighbor knowledge as provided by NHDP
[RFC6130] or from external sources. Unlike S-MPR operation, there is
no need for associating link-layer address information with 1-hop
neighbors since MPR-CDS forwarding is independent of the previous hop
similar to E-CDS forwarding.
To advertise an optional Router Priority value or "WILLINGNESS", an
originating router may use the Message TLV of type SMF_TYPE:MPR-CDS,
which shares a common <value> format with both SMF_TYPE:E-CDS
(Table 14) and SMF_TYPE:S-MPR (Table 16).
MPR-CDS only requires 1-hop knowledge of Router Priority for correct
operation. In the S-MPR phase of MPR-CDS selection, MPRs are
dynamically determined by each router, and selections MUST be
advertised and dynamically updated using NHDP or an equivalent
protocol or mechanism. The <value> field of the SMF_NBR_TYPE:MPR-CDS
type TLV shares a common format with SMF_NBR_TYPE:S-MPR (Table 17) to
convey MPR selection.
C.4. MPR-CDS Selection Algorithm
This section describes an algorithm for the MPR-CDS selection
process. Note that the selection described is with respect to a
specific interface of the router performing selection, and other
router interfaces referenced are reachable from this reference router
interface. An ordered tuple of Router Priority and Router ID is used
in MPR-CDS relay set selection. The Router ID value should be set to
the largest advertised address of a given router; this information is
provided to one-hop neighbors via NHDP by default. Precedence is
given to the Router Priority portion, and the Router ID value is used
as a tiebreaker. The evaluation of this tuple is referred to as
"RtrPri(n)" in the description below where "n" references a specific
router. Note that it is possible that the Router Priority portion
may be optional and the evaluation of "RtrPri()" be solely based upon
the unique Router ID. Since there MUST NOT be any duplicate address
values among SMF routers, a comparison of "RtrPri(n)" between any two
routers will always be an inequality. The following steps, repeated
upon any changes detected within the 1-hop and 2-hop neighborhood,
describe a basic algorithm for conducting MPR-CDS selection for a
router interface "n0":
1. Perform steps 1-8 of Appendix B.4 to select MPRs from the set of
1-hop neighbors of "n0" and notify/update neighbors of
selections.
2. Upon being selected as an MPR (or any change in the set of
routers selecting "n0" as an MPR):
A. If no neighbors have selected "n0" as an MPR, "n0" does not
act as a relay, and no further steps are taken until a change
in neighborhood topology or selection status occurs.
B. Determine the router "n1_max" that has the maximum "RtrPri()"
of all 1-hop neighbors.
C. If "RtrPri(n0)" is greater than "RtrPri(n1_max)", then "n0"
selects itself as a relay for all multicast packets.
D. Else, if "n1_max" has selected "n0" as an MPR, then "0"
selects itself as a relay for all multicast packets.
E. Otherwise, "n0" does not act as a relay.
It is possible to extend this algorithm to consider neighboring SMF
routers that are known to be statically configured for CF (always
relaying). The modification to the above algorithm is to process
such routers as having a maximum possible Router Priority value.
This is the same as the case for participating routers that have been
configured with a S-MPR "WILLINGNESS" value of "WILL_ALWAYS". It is
expected that routers configured for CF and participating in NHDP
would indicate their status with use of the SMF_TYPE TLV type in
their NHDP_HELLO message TLV block. It is important to note,
however, that CF routers will not select MPR routers and therefore
cannot guarantee connectedness.
Author's Address
Joseph Macker (editor)
NRL
Washington, DC 20375
USA
EMail: macker@itd.nrl.navy.mil