draft-ietf-lsvr-bgp-spf-03.xml

<?xml version="1.0" encoding="US-ASCII"?>
<!-- This template is for creating an Internet Draft using xml2rfc,
    which is available here: http://xml.resource.org. -->
<!DOCTYPE rfc SYSTEM "rfc2629.dtd" [
<!-- One method to get references from the online citation libraries.
    There has to be one entity for each item to be referenced.
    An alternate method (rfc include) is described in the references. -->

<!ENTITY RFC1997 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.1997.xml">
<!ENTITY RFC2119 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml">
<!ENTITY RFC2328 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2328.xml">
<!ENTITY RFC2629 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2629.xml">
<!ENTITY RFC3392 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3392.xml">
<!ENTITY RFC3552 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3552.xml">
<!ENTITY RFC4271 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4271.xml">
<!ENTITY RFC4456 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4456.xml">
<!ENTITY RFC4486 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4486.xml">
<!ENTITY RFC4724 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4724.xml">
<!ENTITY RFC4750 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4750.xml">
<!ENTITY RFC4760 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4760.xml">
<!ENTITY RFC4790 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4790.xml">
<!ENTITY RFC4915 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4915.xml">
<!ENTITY RFC5065 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5065.xml">
<!ENTITY RFC5286 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5286.xml">
<!ENTITY RFC5549 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5549.xml">
<!ENTITY RFC5880 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5880.xml">
<!ENTITY RFC7752 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.7752.xml">
<!ENTITY RFC7606 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.7606.xml">
<!ENTITY RFC7938 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.7938.xml">
<!ENTITY RFC8174 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.8174.xml">
<!ENTITY RFC8402 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.8402.xml">
<!ENTITY RFC8405 SYSTEM
  "http://xml.resource.org/public/rfc/bibxml/reference.RFC.8405.xml">
<!ENTITY I-D.ietf-idr-bgpls-segment-routing-epe SYSTEM
  "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.draft-ietf-idr-bgpls-segment-routing-epe-15.xml">
<!ENTITY I-D.ietf-lsvr-applicability SYSTEM
  "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.draft-ietf-lsvr-applicability-00.xml">
]>

<?xml-stylesheet type='text/xsl' href='rfc2629.xslt' ?>
<!-- used by XSLT processors -->
<!-- For a complete list and description of processing instructions (PIs),
    please see http://xml.resource.org/authoring/README.html. -->
<!-- Below are generally applicable Processing Instructions (PIs) that
    most I-Ds might want to use.
    (Here they are set differently than their defaults in xml2rfc v1.32) -->
<?rfc strict="yes" ?>
<!-- give errors regarding ID-nits and DTD validation -->
<!-- control the table of contents (ToC) -->
<?rfc toc="yes"?>
<!-- generate a ToC -->
<?rfc tocdepth="4"?>
<!-- the number of levels of subsections in ToC. default: 3 -->
<!-- control references -->
<?rfc symrefs="yes"?>
<!-- use symbolic references tags, i.e, [RFC2119] instead of [1] -->
<?rfc sortrefs="yes" ?>
<!-- sort the reference entries alphabetically -->
<!-- control vertical white space
    (using these PIs as follows is recommended by the RFC Editor) -->
<?rfc compact="yes" ?>
<!-- do not start each main section on a new page -->
<?rfc subcompact="no" ?>
<!-- keep one blank line between list items -->
<!-- end of list of popular I-D processing instructions -->
<rfc category="std" docName="draft-ietf-lsvr-bgp-spf-03.txt"
 ipr="pre5378Trust200902">
  <!-- category values: std, bcp, info, exp, and historic
    ipr values: full3667, noModification3667, noDerivatives3667
    you can add the attributes updates="NNNN" and obsoletes="NNNN"
    they will automatically be output with "(if approved)" -->

  <!-- ***** FRONT MATTER ***** -->

  <front>

    <title abbrev="BGP Protocol SPF Extensions">
    Shortest Path Routing Extensions for BGP Protocol </title>

    <!-- add 'role="editor"' below for the editors if appropriate -->

    <!-- Another author who claims to be an editor -->

    <author fullname="Keyur Patel" initials="K"
            surname="Patel">

      <organization>Arrcus, Inc.</organization>

      <address>
        <email>keyur@arrcus.com</email>
      </address>
    </author>

    <author fullname="Acee Lindem" initials="A"
            surname="Lindem">

      <organization>Cisco Systems</organization>

      <address>
        <postal>
          <street>301 Midenhall Way</street>

          <city>Cary</city>

          <region>NC</region>

          <code>27513</code>

          <country>USA</country>
        </postal>

        <email>acee@cisco.com</email>

      </address>
    </author>

    <author fullname="Shawn Zandi" initials="S"
            surname="Zandi">

      <organization>Linkedin</organization>

      <address>
        <postal>
          <street>222 2nd Street</street>

          <city>San Francisco</city>

          <region>CA</region>

          <code>94105</code>

          <country>USA</country>
        </postal>

        <email>szandi@linkedin.com</email>

      </address>
    </author>

    <author fullname="Wim Henderickx" initials="W"
            surname="Henderickx">

      <organization>Nokia</organization>

      <address>
        <postal>
          <street></street>

          <city>Antwerp</city>

          <region></region>

          <code></code>

          <country>Belgium</country>
        </postal>

        <email>wim.henderickx@nokia.com</email>

      </address>
    </author>

    <date/>

    <!-- Meta-data Declarations -->

    <area>General</area>

    <workgroup>Network Working Group</workgroup>

    <keyword>IDR</keyword>

    <!-- Keywords will be incorporated into HTML output
        files in a meta tag but they have no effect on text or nroff
        output. If you submit your draft to the RFC Editor, the
        keywords will be used for the search engine. -->

<abstract>
<t>
Many Massively Scaled Data Centers (MSDCs) have converged on simplified 
layer 3 routing. Furthermore, requirements for operational simplicity 
have lead many of these MSDCs to converge on BGP as their single routing
protocol for both their fabric routing and their Data Center Interconnect
(DCI) routing. This document describes a solution which leverages BGP 
Link-State distribution and the Shortest Path First (SPF) algorithm similar
to Internal Gateway Protocols (IGPs) such as OSPF.
</t>
</abstract>

  </front>

  <middle>

<section anchor="introduction" title="Introduction">
<t>
Many Massively Scaled Data Centers (MSDCs) have converged on simplified 
layer 3 routing. Furthermore, requirements for operational simplicity 
have lead many of these MSDCs to converge on BGP <xref target="RFC4271"/> 
as their single routing protocol for both their fabric routing and
their Data Center Interconnect (DCI) routing. Requirements and procedures
for using BGP are described in <xref target="RFC7938"/>.
This document describes an alternative solution which leverages 
BGP-LS <xref target="RFC7752"/> and the Shortest Path First algorithm similar
to Internal Gateway Protocols (IGPs) such as OSPF <xref target="RFC2328"/>. 
</t>
<t>
<xref target="RFC4271"/> defines the 
Decision Process that is used to select routes for subsequent advertisement
by applying the policies in the local Policy Information Base (PIB) to the
routes stored in its Adj-RIBs-In. The output of the Decision Process is the 
set of routes that are announced by a BGP speaker to its peers. These
selected routes are stored by a BGP speaker in the speaker's Adj-RIBs-Out
according to policy.
</t>

<t>
<xref target="RFC7752"/> describes a mechanism by which link-state and TE information can
be collected from networks and shared with external components using BGP.
This is achieved by defining NLRI advertised within the BGP-LS/BGP-LS-SPF
AFI/SAFI. The BGP-LS extensions defined in <xref target="RFC7752"/> makes use of the 
Decision Process defined in <xref target="RFC4271"/>. 
</t>

<t>
This document augments <xref target="RFC7752"/> by replacing its use of the existing 
Decision Process. Rather than reusing the BGP-LS SAFI, the BGP-LS-SPF SAFI
is introduced to insure backward compatibility. The Phase 1 and 2 decision functions of the
Decision Process are replaced with the Shortest Path First (SPF) algorithm
also known as the Dijkstra algorithm. The Phase 3 decision function is also 
simplified since it is no longer dependent on the previous phases. 
This solution avails the benefits of both BGP and SPF-based IGPs.
These include TCP based flow-control, no periodic link-state refresh, and 
completely incremental NLRI advertisement. These advantages can reduce the 
overhead in MSDCs where there is a high degree of Equal Cost Multi-Path 
(ECMPs) and the topology is very stable. 
Additionally, using a SPF-based computation can support fast convergence and 
the computation of Loop-Free Alternatives (LFAs) <xref target="RFC5286"/> in the 
event of link failures. 
Furthermore, a BGP based solution lends itself to multiple peering models
including those incorporating route-reflectors <xref target="RFC4456"/> 
or controllers. 
</t>
<t> 
Support for Multiple Topology Routing (MTR) as described in 
<xref target="RFC4915"/> is an area for further study dependent on deployment 
requirements. 
</t>
<section title="BGP Shortest Path First (SPF) Motivation"> 
<t>
Given that <xref target="RFC7938"/> already describes how BGP could be used
as the sole routing protocol in an MSDC, one might question the motivation for
defining an alternate BGP deployment model when a mature solution exists. 
For both alternatives, BGP offers the operational benefits of a single 
routing protocol. However, BGP SPF offers some unique advantages above 
and beyond standard BGP distance-vector routing.
</t>
<t>
A primary advantage is that all BGP speakers in the BGP SPF routing domain
will have a complete view of the topology. This will allow support for ECMP, 
IP fast-reroute (e.g., Loop-Free Alternatives), Shared Risk Link Groups 
(SRLGs), and other routing enhancements without advertisement of addition 
BGP paths or other extensions. In short, the advantages of an IGP such as 
OSPF <xref target="RFC2328"/> are availed in BGP.
</t>
<t>
With the simplified BGP decision process as defined in <xref target="Phase-1"/>, 
NLRI changes can be disseminated throughout the BGP routing domain much
more rapidly (equivalent to IGPs with the proper implementation).
</t>
<t> 
Another primary advantage is a potential reduction in NLRI advertisement. 
With standard BGP distance-vector routing, a single link failure may impact 
100s or 1000s prefixes and result in the withdrawal or re-advertisement of 
the attendant NLRI. With BGP SPF, only the BGP speakers corresponding to 
the link NLRI need withdraw the corresponding BGP-LS Link NLRI. This 
advantage will contribute to both faster convergence and better scaling. 
</t>
<t>
With controller and route-reflector peering models, BGP SPF advertisement
and distributed computation require a minimal number of sessions and
copies of the NLRI since only the latest version of the NLRI from the 
originator is required. Given that verification of the adjacencies is done
outside of BGP (see <xref target="peering-models"/>), each BGP speaker will 
only need as many sessions and copies of the NLRI as required for
redundancy (e.g., one for the SPF computation and another for backup). 
Functions such as Optimized Route Reflection (ORR) 
are supported without extension by virtue of the primary advantages. 
Additionally, a controller could inject topology that 
is learned outside the BGP routing domain.
</t>
<t>
Given that controllers are already consuming BGP-LS NLRI 
<xref target="RFC7752"/>, reusing for the BGP-LS SPF leverages the 
existing controller implementations. 
</t>
<t> 
Another potential advantage of BGP SPF is that both IPv6 and IPv4 can be 
supported in the same address family using the same topology. Although not
described in this version of the document, multi-topology extensions can
be used to support separate IPv4, IPv6, unicast, and multicast topologies
while sharing the same NLRI. 
</t> 
<t>
Finally, the BGP SPF topology can be used as an underlay for other BGP 
address families (using the existing model) and realize all the above
advantages. A simplified peering model using IPv6 link-local addresses
as next-hops can be deployed similar to  <xref target="RFC5549"/>.
</t>
</section>
      <section title="Requirements Language">
       <t>The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
        NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED",
        "MAY", and "OPTIONAL" in this document are to be interpreted as
        described in BCP 14 <xref target="RFC2119"/> <xref target="RFC8174"/> 
        when, and only when, they appear in all capitals, as shown here.</t>
       </section>
</section> <!-- for Introductions section -->


<section anchor="peering-models" title="BGP Peering Models">
<t>
Depending on the requirements, scaling, and capabilities of the BGP speakers, various
peering models are supported. The only requirement is that all BGP speakers in the
BGP SPF routing domain receive link-state NLRI on a timely basis, run an SPF calculation, 
and update their data plane appropriately. The content of the Link NLRI is described 
in <xref target="Link-NLRI"/>.
</t>
<section title="BGP Single-Hop Peering on Network Node Connections">
<t>
The simplest peering model is the one described in section 5.2.1 of 
<xref target="RFC7938"/>. In this model,
EBGP single-hop sessions are established over direct point-to-point links
interconnecting the SPF domain nodes. For the purposes of BGP SPF, Link NLRI is only 
advertised if a single-hop BGP session has been established and the Link-State/SPF
address family capability has been exchanged <xref target="RFC4790"/> on the
corresponding session. 
If the session goes down, the corresponding Link NLRI will be withdrawn. Topologically,
this would be equivalent to the peering model in <xref target="RFC7938"/> where there
is a BGP session on every link in the data center switch fabric.
</t>
</section>
<section title="BGP Peering Between Directly Connected Network Nodes">
<t>
In this model, BGP speakers peer with all directly connected
network nodes but the sessions may be multi-hop and the direct connection
discovery and liveliness detection for those connections are 
independent of the BGP protocol. How this is accomplished is outside
the scope of this document. 
Consequently, there will be a single session even if there are multiple 
direct connections between BGP speakers.
For the purposes of BGP SPF, Link NLRI is advertised as long as 
a BGP session has been established, the Link-State/SPF address family 
capability has been exchanged <xref target="RFC4790"/> and
the corresponding link is considered is up and considered operational.
This is much like the previous peering model only peering is on a single
loopback address and the switch fabric links can be unnumbered. However,
there will be the same unnumber of sessions as with the previous peering
model unless there are parrallel links between switches in the fabric.
</t>
</section>
<section title="BGP Peering in Route-Reflector or Controller Topology">
<t>
In this model, BGP speakers peer solely with one or more Route Reflectors
<xref target="RFC4456"/> or controllers. As in the previous model, direct
connection discovery and liveliness detection for those connections are 
done outside the BGP protocol. More specifically, the Liveliness detection is
done using BFD protocol described in <xref target="RFC5880"/>. For the 
purposes of BGP SPF, Link NLRI is 
advertised as long as the corresponding link is up and considered operational. 
</t>
<t>This peering model, known as sparse peering, allows for many fewer BGP sessions 
and, consequently, instances of the same NLRI received from multiple peers. It is
discussed in greater detail in <xref target="I-D.ietf-lsvr-applicability"/>. 
</t>
</section>
</section>

<section title="BGP-LS Shortest Path Routing (SPF) SAFI">
<t>
  In order to replace the Phase 1 and 2 decision functions of the 
  existing Decision Process with an SPF-based Decision Process and streamline 
  the Phase 3 decision functions in a backward compatible manner, this draft 
  introduces the BGP-LS-SFP SAFI for BGP-LS SPF operation.
  The BGP-LS-SPF (AF 16388 / SAFI TBD1) <xref target="RFC4790"/> is allocated by IANA
  as specified in the <xref target="IANA"/>. A BGP speaker using the
  BGP-LS SPF extensions described herein MUST exchange the AFI/SAFI using 
  Multiprotocol Extensions Capability Code <xref target="RFC4760"/> with 
  other BGP speakers in the SPF routing domain.
</t>
</section>

<section title="Extensions to BGP-LS">
<t>
  <xref target="RFC7752"/> describes a mechanism by which link-state and TE
  information can be collected from networks and shared with external components
  using BGP protocol. It describes both the definition of BGP-LS NLRI
   that describes links, nodes, and prefixes comprising IGP link-state
   information and the definition of a BGP path attribute (BGP-LS
   attribute) that carries link, node, and prefix properties and
   attributes, such as the link and prefix metric or auxiliary 
   Router-IDs of nodes, etc.
</t>

<t>
The BGP protocol will be used in the Protocol-ID field specified in
table 1 of <xref target="I-D.ietf-idr-bgpls-segment-routing-epe"/>. 
The local and remote node descriptors for all NLRI will be the BGP Router-ID (TLV 516) 
and either the AS Number (TLV 512) <xref target="RFC7752"/> or the BGP Confederation
Member (TLV 517) <xref target="RFC8402"/>. 
However, if the BGP Router-ID is known to be unique within the BGP Routing domain,
it can be used as the sole descriptor. 
</t>
<section title="Node NLRI Usage and Modifications">
<t>
The SPF capability is a new Node Attribute TLV that will be added
to those defined in table 7 of <xref target="RFC7752"/>. The
new attribute TLV will only be applicable when BGP is specified 
in the Node NLRI Protocol ID field. 
The TBD TLV type will be defined by IANA. The new Node
Attribute TLV will contain a single-octet SPF algorithm as defined 
in <xref target="RFC8402"/>.
</t>
<t>
<figure align="center">
<artwork align="left"><![CDATA[
    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
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |              Type             |             Length            |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   | SPF Algorithm |
   +-+-+-+-+-+-+-+-+

The SPF Algorithm may take the following values:

0 - Normal Shortest Path First (SPF) algorithm based on link
    metric. This is the standard shortest path algorithm as
    computed by the IGP protocol.  Consistent with the deployed
    practice for link-state protocols, Algorithm 0 permits any
    node to overwrite the SPF path with a different path based on
    its local policy.
1 - Strict Shortest Path First (SPF) algorithm based on link
    metric. The algorithm is identical to Algorithm 0 but Algorithm
    1 requires that all nodes along the path will honor the SPF
    routing decision.  Local policy at the node claiming support for
    Algorithm 1 MUST NOT alter the SPF paths computed by Algorithm 1.
]]></artwork>
</figure>
</t>
<t>Note that usage of Strict Shortest Path First (SPF) algorithm is
defined in the IGP algorithm registry but usage is restricted to 
<xref target="I-D.ietf-idr-bgpls-segment-routing-epe"/>. Hence, its 
usage for BGP-LS SPF is out of scope.</t>
<t> 
When computing the SPF for a given BGP routing domain, only BGP nodes
advertising the SPF capability attribute will be included the Shortest
Path Tree (SPT). 
</t>
</section>
<section anchor="Link-NLRI" title="Link NLRI Usage">
<t>
The criteria for advertisement of Link NLRI are discussed in 
<xref target="peering-models"/>.
</t>
<t>
Link NLRI is advertised with local and remote node descriptors as described 
above and unique link identifiers dependent on the addressing. For IPv4 links, the 
links local IPv4 (TLV 259) and remote IPv4 (TLV 260) addresses will be used. 
For IPv6 links, the local IPv6 (TLV 261) and remote IPv6 (TLV 262) addresses
will be used. For unnumbered links, the link local/remote identifiers (TLV 258)
will be used. For links supporting having both IPv4 and IPv6 addresses, both sets
of descriptors may be included in the same Link NLRI. The link identifiers are 
described in table 5 of <xref target="RFC7752"/>.
</t>
<t>
The link IGP metric attribute TLV (TLV 1095) as well as any others required for non-SPF 
purposes SHOULD be advertised.
Algorithms such as setting the metric inversely to the link speed as done in the 
OSPF MIB <xref target="RFC4750"/> MAY be supported. However, this is beyond the scope 
of this document. 
</t>
<section title="BGP-LS Link NLRI Attribute Prefix-Length TLVs">
<t>
Two BGP-LS Attribute TLVs to BGP-LS Link NLRI are defined to advertise the prefix length
associated with the IPv4 and IPv6 link prefixes. The prefix length is used for the
optional installation of prefixes corresponding to Link NLRI as defined in 
<xref target="BGP-SPF"/>.</t>
<t>
<figure align="center">
<artwork align="left"><![CDATA[
    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
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |      TBD IPv4 or IPv6 Type    |             Length            |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   | Prefix-Length |
   +-+-+-+-+-+-+-+-+

     Prefix-length - A one-octet length restricted to 1-32 for IPv4 
                     Link NLIR endpoint prefixes and 1-128 for IPv6
                     Link NLRI endpoint prefixes.
]]></artwork>
</figure>
</t>
</section>
</section>
<section title="Prefix NLRI Usage">
<t>
Prefix NLRI is advertised with a local node descriptor as described above and the prefix and
length used as the descriptors (TLV 265) as described in <xref target="RFC7752"/>. 
The prefix metric attribute TLV (TLV 1155) as well as any others required for non-SPF 
purposes SHOULD be advertised. For loopback prefixes, the metric should be 0. For non-loopback 
prefixes, the setting of the metric is a local matter and beyond the scope of this document. 
</t>
</section>
<section title="BGP-LS Attribute Sequence-Number TLV">
<t>
A new BGP-LS Attribute TLV to BGP-LS NLRI types is defined to assure the most
recent version of a given NLRI is used in the SPF computation. 
The TBD TLV type will be defined by IANA. The new BGP-LS
Attribute TLV will contain an 8-octet sequence number. The usage of the Sequence Number TLV
is described in <xref target="Phase-1"/>. 
</t>
<t>
<figure align="center">
<artwork align="left"><![CDATA[
    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
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |              Type             |             Length            |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                Sequence Number (High-Order 32 Bits)           | 
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                Sequence Number (Low-Order 32 Bits)            | 
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
]]></artwork>
</figure>
</t>
<t>  
Sequence Number <vspace blankLines="1" />		
The 64-bit strictly increasing sequence number is incremented for every 
version of BGP-LS NLRI originated. BGP speakers implementing this specification MUST use 
available mechanisms to preserve the sequence number's strictly increasing property 
for the deployed life of the BGP speaker (including cold restarts). 
One mechanism for accomplishing this would be to use the high-order 32 bits of the 
sequence number as a wrap/boot count that is incremented anytime the BGP router
loses its sequence number state or the low-order 32 bits wrap.
</t>
<t>
When incrementing the sequence number for each self-originated NLRI, 
the sequence number should be treated as an unsigned 64-bit
value. If the lower-order 32-bit value wraps, the higher-order 32&nbhy;bit value should
be incremented and saved in non-volatile storage. If by some chance the BGP Speaker is 
deployed long enough that there is a possibility that the 64-bit sequence number may wrap 
or a BGP Speaker completely loses its sequence number state (e.g., the BGP speaker hardware
is replaced or experiences a cold-start), the phase 1 decision function 
(see <xref target="Phase-1"/>) rules will insure convergence, albeit, not immediately.
</t>
</section>
</section> 

<section title="Decision Process with SPF Algorithm">
<t>
The Decision Process described in <xref target="RFC4271"/> takes place in 
three distinct phases. The Phase 1 decision function of the Decision Process is 
responsible for calculating the degree
of preference for each route received from a BGP speaker's peer. The Phase 2 decision
function is invoked on completion of the Phase 1 decision function and is responsible 
for choosing the best route out of all those available for each
distinct destination, and for installing each chosen route into the Loc-RIB. 
The combination of the Phase 1 and 2 decision functions is characterized as 
a Path Vector algorithm.
</t>
<t>
The SPF based Decision process replaces the BGP best-path Decision process described in 
<xref target="RFC4271"/>. This process starts with selecting only those Node NLRI whose 
SPF capability TLV matches with the local BGP speaker's SPF capability TLV value. 
Since Link-State NLRI always contains the local descriptor 
<xref target="RFC7752"/>, it will only 
be originated by a single BGP speaker in the BGP routing domain. 
These selected Node NLRI and their Link/Prefix NLRI are used to build a directed
graph during the SPF computation. The best paths for BGP prefixes
are installed as a result of the SPF process. 
</t>
<t>
When BGP-LS-SPF NLRI is received, all that is required is to determine 
whether it is the best-path by examining the Node-ID and sequence number as described
in <xref target="Phase-1"/>. If the received best-path NLRI had changed, it will be advertised
to other BGP-LS-SPF peers. If the attributes have changed (other than the sequence number), 
a BGP SPF calculation will be scheduled. However, a changed NLRI MAY be 
advertised to other peers almost immediately and propagation of changes can approach 
IGP convergence times.  To accomplish this, the MinRouteAdvertisementIntervalTimer and
MinASOriginationIntervalTimer <xref target="RFC4271"/> are not applicable
to the BGP-LS-SPF SAFI. Rather, SPF calculations SHOULD be triggered and dampened consistent
with the SPF backoff algorithm specified in <xref target="RFC8405"/>.
</t>
<t>
The Phase 3 decision function 
of the Decision Process <xref target="RFC4271"/> is also simplified since under
normal SPF operation, a BGP speaker would advertise the NLRI
selected for the SPF to all BGP peers with the BGP-LS/BGP-LS-SPF AFI/SAFI. 
Application of policy would not be prevented however its usage to best-path process
would be limited as the SPF relies solely on link metrics.  
</t>

<section anchor="Phase-1" title="Phase-1 BGP NLRI Selection">
<t>
The rules for NLRI selection are greatly simplified from <xref target="RFC4271"/>.
<list style="numbers">
<t>
If the NLRI is received from the BGP speaker originating the NLRI (as determined by the
comparing BGP Router ID in the NLRI Node identifiers with the BGP speaker Router ID), 
then it is preferred over the same NLRI from non-originators. This rule will assure that
stale NLRI is updated even if a BGP-LS router loses its sequence number state due to a
cold-start.
</t>
<t>
If the Sequence-Number TLV is present in the BGP-LS Attribute, then the NLRI with the
most recent, i.e., highest sequence number is selected. BGP-LS NLRI with a Sequence-Number
TLV will be considered more recent than NLRI without a BGP-LS Attribute or a 
BGP-LS Attribute that doesn't include the Sequence-Number TLV. 
</t>
<t>The final tie-breaker is the NLRI from the BGP Speaker with the numerically largest
BGP Router ID.
</t>
</list> 
</t>
<t>
When a BGP speaker completely loses its sequence number state, i.e., due to a cold start, or
in the unlikely possibility that that sequence number wraps, the BGP routing domain will
still converge. This is due to the fact that BGP speakers adjacent to the router will
always accept self-originated NLRI from the associated speaker as more recent (rule # 1). When
BGP speaker reestablishes a connection with its peers, any existing session will be taken
down and stale NLRI will be replaced by the new NLRI and stale NLRI will be discarded 
independent of whether or not BGP graceful restart is deployed, <xref target="RFC4724"/>. The adjacent
BGP speaker will update their NLRI advertisements in turn until the BGP routing domain has
converged. 
</t>
<t>
The modified SPF Decision Process performs an SPF calculation rooted at the BGP 
speaker using the metrics from
Link and Prefix NLRI Attribute TLVs <xref target="RFC7752"/>. As a result, any attributes that
would influence the Decision process defined in <xref target="RFC4271"/> like ORIGIN, MULTI_EXIT_DISC, and 
LOCAL_PREF attributes are ignored by the SPF algorithm. Furthermore, the NEXT_HOP attribute
value is preserved but otherwise ignored during the SPF or best-path. 
</t>

</section>

<section title="Dual Stack Support">
<t>
The SPF-based decision process operates on Node, Link, and Prefix NLRIs that support
both IPv4 and IPv6 addresses. Whether to run a single SPF instance or multiple
SPF instances for separate AFs is a matter of a local implementation. Normally, IPv4
next-hops are calculated for IPv4 prefixes and IPv6 next-hops are calculated for IPv6
prefixes. However, an interesting use-case is deployment of <xref target="RFC5549"/> where 
IPv6 next-hops are calculated for both IPv4 and IPv6 prefixes. As stated in 
<xref target="introduction"/>, support for Multiple Topology Routing (MTR) is an area 
for future study.
</t>
</section>

<section anchor="BGP-SPF" title="SPF Calculation based on BGP-LS NLRI">
<t>This section details the BGP-LS SPF local routing information base (RIB) calculation.
The router will use BGP-LS Node, Link, and Prefix NLRI to populate the local RIB using the
following algorithm. This calculation yields the set of intra-area routes associated
with the BGP-LS domain.  A router calculates the  shortest-path tree using itself
as the root. Variations and optimizations of the algorithm are valid as long as it
yields the same set of routes. The algorithm below supports Equal Cost Multi-Path (ECMP)
routes. Weighted Unequal Cost Multi-Path are out of scope. The organization of this section
owes heavily to section 16 of <xref target="RFC2328"/>.</t>
<t>The following abstract data structures are defined in order to specify the algorithm.
<list style="symbols">
<t>Local Route Information Base (RIB) - This is abstract contains reachability information 
(i.e., next hops) for all prefixes (both IPv4 and IPv6) as well as the Node NLRI 
reachability. Implementations may choose to implement this as separate RIBs for each
address family and/or Node NLRI.</t>
<t>Link State NLRI Database (LSNDB) - Database of BGP-LS NLRI that facilitates access to 
all Node, Link, and Prefix NLRI as well as all the Link and Prefix NLRI corresponding to
a given Node NLRI. Other optimization, such as, resolving bi-directional connectivity
associations between Link NLRI are possible but of scope of this document.</t>
<t>Candidate List - This is a list of candidate Node NLRI with the lowest cost Node NLRI
at the front of the list. It is typically implemented as a heap but other concrete
data structures have also been used.</t>
</list></t>
<t>The algorithm is comprised of the steps below:
<list style="numbers">
<t>The current local RIB is invalidated. The local RIB is
built again from scratch.  The existing routing entries are preserved for comparision to
determine changes that need to be installed in the global RIB.</t>
<t>The computing router's Node NLRI is installed in the local RIB with a cost of 0 and as
 as the sole entry in the candidate list.</t>
<t>The Node NLRI with the lowest cost is removed from the candidate list for processing. The
Node corresponding to this NLRI will be referred to as the Current Node. If the candidate
list is empty, the SPF calculation has completed and the algorithm proceeds to step 6.</t>
<t>All the Prefix NLRI with the same Node Identifiers as the Current Node will be considered
for installation. The cost for each prefix is the metric advertised in the Prefix NLRI 
added to the cost to reach the Current Node.
<list style="symbols">
<t>If the prefix is not in the local RIB, the prefix is installed and will inherit the
Current Node's next hops.</t>
<t>If the prefix is in the local RIB and the cost is greater than the Current route's metric,
the Prefix NLRI does not contribute to the route and is ignored.</t>
<t>If the prefix is in the local RIB and the cost is less than the current route's metric, 
the Prefix is installed with the Current Node's next-hops replacing the local RIB route's 
next-hops and the metric being updated.</t>
<t>If the prefix is in the local RIB and the cost is same as the current route's metric, 
the Prefix is installed with the Current Node's next-hops being merged with local 
RIB route's next-hops.</t>
</list> </t>
<t>All the Link NLRI with the same Node Identifiers as the Current Node will be considered
for installation. Each link will be examined and will be referred to in the following text
as the Current Link. The cost of the Current Link is the advertised metric in the Link NLRI 
added to the cost to reach the Current Node. 
<list style="symbols">
<t>Optionally, the prefix(es) associated with the Current Link are installed into the
local RIB using the same rules as were used for Prefix NLRI in the previous steps.</t>
<t>The Current Link's endpoint Node NLRI is accessed (i.e., the Node NLRI
with the same Node identifiers as the Link endpoint). If it exists, it will be referred to 
as the Endpoint Node NLRI and the algorithm will proceed as follows:
<list style="symbols">
<t>All the Link NLRI corresponding the Endpoint Node NLRI will be searched for a back-link 
NLRI pointing to the current node. Both the Node identifiers and the 
Link endpoint identifiers in the Endpoint Node's Link NLRI must match for a match. If there
is no corresponding Link NLRI corresponding to the Endpoint Node NLRI, the Endpoint Node
NLIR fails the bi-directional connectivity test and is not processed further.</t>
<t>If the Endpoint Node NLRI is not on the candidate list, it is inserted based on the 
link cost and BGP Identifier (the latter being used as a tie-breaker).</t>
<t>If the Endpoint Node NLRI is already on the candidate list with a lower cost, it need
not be inserted again.</t>
<t>If the Endpoint Node NLRI is already on the candidate list with a higher cost, it must
be removed and reinserted with a lower cost.</t>
</list></t>
<t>Return to step 3 to process the next lowest cost Node NLRI on the candidate list.</t>
</list></t>
<t>The local RIB is examined and changes (adds, deletes, modifications) are installed into
the global RIB.</t>
</list> 
</t>

</section> 

<section title="NEXT_HOP Manipulation">
<t>
A BGP speaker that supports SPF extensions MAY interact with peers that don't support
SPF extensions. If the BGP-LS address family is advertised to a peer not
supporting the SPF extensions described herein, then the BGP speaker 
MUST conform to the NEXT_HOP rules specified in <xref target="RFC4271"/> when announcing 
the Link-State address family routes to those peers.
</t>

<t>
All BGP peers that support SPF extensions would locally compute the Loc-RIB next-hops 
as a result of the SPF process. Consequently, the NEXT_HOP attribute is always ignored on 
receipt. However, BGP speakers SHOULD set the NEXT_HOP address according to the
NEXT_HOP attribute rules specified in <xref target="RFC4271"/>.
</t>
</section>

<section title="IPv4/IPv6 Unicast Address Family Interaction">
<t>
While the BGP-LS SPF address family and the IPv4/IPv6 unicast address families install routes 
into the same device routing tables, they will operate independently much the same as OSPF and
IS-IS would operate today (i.e., "Ships-in-the-Night" mode). There will be no implicit
route redistribution between the BGP address families. However, implementation specific
redistribution mechanisms SHOULD be made available with the restriction that redistribution
of BGP-LS SPF routes into the IPv4 address family applies only to IPv4 routes and redistribution
of BGP-LS SPF route into the IPv6 address family applies only to IPv6 routes.
</t>
<t> 
Given the fact that SPF algorithms are based on the assumption that all routers in the
routing domain calculate the precisely the same SPF tree and install the same set of
routes, it is RECOMMENDED that BGP-LS SPF IPv4/IPv6 routes be given priority by default
when installed into their respective RIBs. In common implementations the prioritization
is governed by route preference or administrative distance with lower being more preferred.
</t>
</section>


<section anchor="NLRI-Advertise" title="NLRI Advertisement and Convergence">
<t>A local failure will prevent a link from being used in the SPF calculation 
due to the IGP bi-directional connectivity requirement. Consequently, local link
failures should always be given priority over updates (e.g., withdrawing all 
routes learned on a session) in order to ensure the highest priority propagation 
and optimal convergence.
</t>
<t>Delaying the withdrawal of non-local routes is an area for further study as 
more IGP-like mechanisms would be required to prevent usage of stale NLRI.</t>
</section> 

<section title="Error Handling">
<t>
When a BGP speaker receives a BGP Update containing a malformed SPF Capability TLV
in the Node NLRI BGP-LS Attribute <xref target="RFC7752"/>,
it MUST ignore the received TLV and the Node NLRI and not pass it to other BGP peers as
specified in <xref target="RFC7606"/>.
When discarding a Node NLRI with malformed TLV, a BGP speaker SHOULD log an error for 
further analysis. 
</t>

</section>
</section>

    <section anchor="IANA" title="IANA Considerations">
<t>
  This document defines an AFI/SAFI for BGP-LS SPF operation and
  requests IANA to assign the BGP-LS/BGP-LS-SPF (AFI 16388 / SAFI TBD1)
  as described in <xref target="RFC4750"/>.
</t>
<t>
  This document also defines four attribute TLVs for BGP LS NLRI.
  We request IANA to assign TLVs for the SPF capability, 
  Sequence Number, IPv4 Link Prefix-Length, and IPv6 Link Prefix-Length
  from the "BGP-LS Node Descriptor, Link Descriptor,  Prefix Descriptor, 
  and  Attribute TLVs" Registry. 
</t>
</section>

<section anchor="Security" title="Security Considerations">
<t>
This extension to BGP does not change the underlying security issues
inherent in the existing <xref target="RFC4271"/>, <xref target="RFC4724"/>,
and <xref target="RFC7752"/>.
</t>
</section>

<section anchor="Management" title="Management Considerations">
<t>
This section includes unique management considerations for the BGP-LS SPF address family. 
</t>
<section anchor="Config" title="Configuration">
<t>
In addition to configuration of the BGP-LS SPF address family, implementations SHOULD 
support the configuratio of the INITIAL_SPF_DELAY, SHORT_SPF_DELAY, LONG_SPF_DELAY,
TIME_TO_LEARN, and HOLDDOWN_INTERVAL as documented in <xref target="RFC8405"/>.
</t>
</section>
<section anchor="Operation" title="Operational Data">
<t>
In order to troubleshoot SPF issues, implementations SHOULD support an SPF log including
entries for previous SPF computations, Each SPF log entry would include the BGP-LS NLRI SPF 
triggering the SPF, SPF scheduled time, SPF start time, SPF end time, and SPF type if 
different types of SPF are supported. Since the size of the log will be finite, implementations 
SHOULD also maintain counters for the total number of SPF computations of each type and the
total number of SPF triggering events. Additionally, to troubleshoot SPF scheduling and
backoff <xref target="RFC8405"/>, the current SPF backoff state, remaining time-to-learn, 
remaining holddown, last trigger event time, last SPF time, and next SPF time should be 
available.
</t> 
</section>
</section>

<section anchor="Acknowledgements" title="Acknowledgements">
<t>
The authors would like to thank Sue Hares, Jorge Rabadan, Boris Hassanov, Dan Frost, 
and Fred Baker for their review and comments.
</t>
</section>

<section anchor="Contributors" title="Contributors">
<t>
In addition to the authors listed on the front page, the following
co-authors have contributed to the document.
</t>

<figure align="center">
<artwork align="left"><![CDATA[
  Derek Yeung
  Arrcus, Inc.
  derek@arrcus.com

  Gunter Van De Velde
  Nokia
  gunter.van_de_velde@nokia.com

  Abhay Roy
  Cisco Systems
  akr@cisco.com

  Venu Venugopal
  Cisco Systems
  venuv@cisco.com
]]></artwork>
</figure>

</section>


  </middle>

  <!--  *****BACK MATTER ***** -->

  <back>
    <!-- References split into informative and normative -->

    <!-- There are 2 ways to insert reference entries from the citation libraries:
    1. define an ENTITY at the top, and use "ampersand character"RFC2629;
        here (as shown)
    2. simply use a PI
        "less than character"?rfc include="reference.RFC.2119.xml"?> here
        (for I-Ds:
          include="reference.I-D.narten-iana-considerations-rfc2434bis.xml")

    Both are cited textually in the same manner: by using xref elements.
    If you use the PI option, xml2rfc will, by default, try to find included
    files in the same directory as the including file. You can also define
    the XML_LIBRARY environment variable
    with a value containing a set of directories to search.  These can be
    either in the local
    filing system or remote ones accessed by http (http://domain/dir/... ).-->

    <references title="Normative References">

      &RFC2119;

      &RFC4271;

      &RFC7606;

      &RFC7752;

      &RFC7938;

      &RFC8174;

      &RFC8402;

      &RFC8405;

      &I-D.ietf-idr-bgpls-segment-routing-epe;

    </references>

    <references title="Information References">

      &RFC2328;

      &RFC4456;

      &RFC4724;

      &RFC4750;

      &RFC4760;

      &RFC4790;

      &RFC4915;

      &RFC5286;

      &RFC5549;

      &RFC5880;

      &I-D.ietf-lsvr-applicability;
    </references>

    <!-- Change Log

v00 2008-10-01  KP    Initial version
    -->
  </back>
</rfc>