One document matched: draft-ward-irs-framework-00.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 I-D.ietf-pce-stateful-pce SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-pce-stateful-pce.xml">

<!ENTITY I-D.ietf-isis-genapp SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ietf-isis-genapp.xml">

<!ENTITY I-D.gredler-idr-ls-distribution SYSTEM "http://xml.resource.org/public/rfc/bibxml3/reference.I-D.gredler-idr-ls-distribution.xml">
<!ENTITY RFC3137 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3137.xml">
<!ENTITY RFC5250 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5250.xml">
<!ENTITY RFC5470 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5470.xml">
<!ENTITY RFC5575 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5575.xml">
<!ENTITY RFC5693 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5693.xml">

<!ENTITY RFC6020 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.6020.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="no" ?>
<!-- give errors regarding ID-nits and DTD validation -->
<!-- control the table of contents (ToC) -->
<?rfc toc="yes"?>
<!-- generate a ToC -->
<?rfc tocdepth="5"?>
<!-- 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="info" docName="draft-ward-irs-framework-00" ipr="trust200902">
  <!-- 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>
    <!-- The abbreviated title is used in the page header - it is only necessary if the 
         full title is longer than 39 characters -->

    <title abbrev="IRS Framework">Interface to the Routing System Framework</title>

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

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

    <author fullname="Alia Atlas" initials="A.K.A." surname="Atlas" role="editor">
     <organization>Juniper Networks</organization>
     <address>
       <postal>
         <street>10 Technology Park Drive</street>
         <city>Westford</city>
         <region>MA</region>
         <code>01886</code>
         <country>USA</country>
       </postal>
       <email>akatlas@juniper.net</email>
      </address>
    </author>

    <author fullname="Thomas Nadeau" initials="T.N." surname="Nadeau">
     <organization>Juniper Networks</organization>
     <address>
       <postal>
         <street>1194 N. Mathilda Ave.</street>
         <city>Sunnyvale</city>
         <region>CA</region>
         <code>94089</code>
         <country>USA</country>
       </postal>
       <email>tnadeau@juniper.net</email>
      </address>
    </author>

    <author fullname="Dave Ward" initials="D.W." surname="Ward">
      <organization>Cisco Systems</organization>
      <address>
       <postal>
         <street>Tasman Drive</street>
         <city>San Jose</city>
         <region>CA</region>
         <code>95134</code>
         <country>USA</country>
       </postal>
        <email>wardd@cisco.com</email>
      </address>
    </author>

    <date year="2012" />

    <!-- If the month and year are both specified and are the current
    ones, xml2rfc will fill in the current day for you. If only the
    current year is specified, xml2rfc will fill in the current day
    and month for you. If the year is not the current one, it is
    necessary to specify at least a month (xml2rfc assumes day="1" if
    not specified for the purpose of calculating the expiry date).
    With drafts it is normally sufficient to specify just the
    year. -->

    <!-- Meta-data Declarations -->

    <area>Routing</area>

<!--
    <workgroup>Routing Area Working Group</workgroup>
 -->

    <abstract>

      <t>This document describes a framework for a standard,
      programmatic interface for full-duplex, streaming state transfer
      in and out of the Internet's routing system.  It lists the
      information that might be exchanged over the interface, and
      describes the uses of an interface to the Internet routing
      system.</t>
    </abstract>
  </front>

  <middle>
<section anchor="Intro" title="Introduction">

<t>Routers that form the Internet's routing infrastructure maintain
state at various layers of detail and function.  For example, each
router has a Routing Information Base (RIB), and the routing protocols
(OSPF, ISIS, BGP, etc.) each maintain protocol state and information
about the state of the network.</t>

<t>A router also has information that may be required for applications
to understand the network, verify that programmed state is installed
in the forwarding plane, measure the behavior of various flows, and
understand the existing configuration and state of the router.
Furthermore, routers are configured or implemented with procedural or
policy-based instructions for how to convert all of this information
into the forwarding operations that are installed in the forwarding
plane, and this is also state information that describes the behaviour
of the router.</t>

<t>This document sets out a framework for a common, standard interface
to allow access to all of this information.  This Interface to the
Routing System (IRS) would facilitate control and diagnosis of the
routing infrastructure, as well as enabling sophisticated applications
to be built on top of today's routed networks.  The IRS is a
programmatic, streaming interface for transferring state into and out
of the Internet's routing system, and recognizes that the routing
system and a router's OS provide useful mechanisms that applications
could harness to accomplish application-level goals.</t>

<t>Fundamental to the IRS is a clear data model that defines the
semantics of the information that can be written and read.  The IRS
provides a framework for registering for and requesting the
appropriate information for each particular application.  The IRS
provides a way for applications to customize network behaviour while
leveraging the existing routing system.</t>

<t>The IRS, and therefore this document, is specifically focused on an
interface for routing and forwarding data.</t>

<section title="Functional Overview">

<t>There are three key aspects to the IRS.  First, the interface is a
programmatic streaming interface meaning that it is asynchronous and
offers fast, interactive access.Second, the IRS gives access to
information and state that is not usually configurable or modeled in
existing implementations or configuration protocols.  Third, the IRS
gives applications the ability to learn additional, structured,
filterable information and events from the router.</t>

  <t>IRS is described as a streaming programmatic interface; the key
  properties that are intended are:
  <list style="hanging">

    <t hangText="Multiple Simultaneous Asynchronous Operations: ">A
    single application should be able to send multiple operations to
    IRS without needing to wait for each to complete before sending
    the next.</t>

    <t hangText="Configuration Not Re-Processed: ">When an IRS
    operation is processed, it does not require that any of the
    configuration be processed.  I.e. the desired behavior with regard
    to static configuration is the same as learning a new BGP route -
    completely orthogonal.</t>

    <t hangText="Duplex: ">Communications can be established by either
    the router or the application.  Similarly, events,
    acknowledgements, failures, operations, etc. can be sent at any
    time by both the router and the application.  This is not a pure
    pull-model where only the application queries to pull responses.</t>
 
    <t hangText="High-Throughput: ">At a minimum, the IRS should be
    able to handle hundreds of operations per second.</t>

    <t hangText="Responsive: ">It should be possible to complete
    simple operations within a sub-second time-scale.</t>

    <t hangText="Multi-Channel: ">It should be possible for
    information to be communicated via the interface from different
    components in the router without requiring going through a single
    channel.  For example, for scaling, some exported data or events
    may be better sent directly from the forwarding plane, while other
    interactions may come from the control-plane.  Thus a single TCP
    session per application would not be a good match.</t>

  </list> </t>

  <t>Such an interface facilitates the specification of non-permanent
  state into the routing system as well as the extraction of that
  information and additional dynamic information from the routing
  system.  A non-routing protocol or application could inject state
  into a networking node's OS via the state-insertion aspects of the
  interface, that could then be distributed in a routing or signaling
  protocol.</t>

  <t>Where existing mechanisms can provide part of the desired
  functionality, the coverage and gaps are briefly discussed in this
  document.</t>

  <t>The existing mechanisms, such as SNMP and NetConf, that allow
  state to be written and read do not meet all of the above key
  properties needed for IRS.  The overhead of infrastructure is also
  quite high and many MIBs do not, in definition or practice, allow
  writing of state.  There is also very limited capability to add new
  application-specific state to be distributed via the routing system.
  Conversely, NetConf is challenging for reading state from a
  router.</t>

  <t>ForCES is another method for writing state into a router, but its
  focus is on the forwarding plane.  By focusing on the forwarding
  plane, it requires that the forwarding plane be modeled and
  programmable and ignores the existence and intelligence of the
  router OS and routing system.  ForCES provides a lower-level
  interface than IRS is intended to address.</t>

</section>
<section title="Example Use-Cases">

<t>A few brief examples of ways an application could use the IRS are
presented here.  These are intended to give a sense of what could be
done rather than to be primary and detailed motivational
use-cases.</t>

<t><list style="hanging">

<t hangText="Route Control via Indirection: ">By enabling an
application to install routes in the RIB, it is possible that when,
for example, BGP resolves its IGP next-hop via the RIB, that could be
to an application-installed route.  In general, when a route is
redistributed from one protocol to another, this is done via the RIB
and such a route could have been installed via the IRS interface.</t>

<t hangText="Policy-Based Routing of Unknown Traffic: ">A static
route, installed into the RIB, could direct otherwise unrecognized
traffic towards an application, through whatever appropriate tunnel
was required, for further handling.  Such a static route could be
programmed with indirection, so that its outgoing path is whatever is
used by another particular route (e.g. to a particular server).</t>

<t hangText="Services with Fixed Hours: ">If an application were to
provide services only during fixed time-periods, the application could
install both a specific route on the local router in the RIB and
advertise the associated prefix as being attached to the local router
via the IGP.  If the application knew the fixed hours, the state so
installed could be time-based and automatically removed at
approximately the correct time.</t>

<t hangText="Traffic Mirroring:  ">The interface to the multicast RIB
could be used to mirror a particular traffic flow to both its original
destination and a data collector.</t>

<t hangText="Static Multicast Trees: ">An application could set up
static (or partially static) multicast flows via entries in the
multicast RIB without requiring an associated multicast protocol.
This could be useful in networks with a fixed topology and
well-planned distribution tree that provides redundancy.</t>

</list></t>
</section>

</section><!-- End of Introduction !-->

<section title="Programmatic Interfaces">

<t>A number of management interfaces exist today that allow for the
indirect programming of the routing system. These include proprietary
CLI, Netconf, and SNMP.  However, none of these mechanisms allows for
the direct programming of the routing system.  Such streaming
interfaces are needed to support dynamic time-based applications.</t>

<t>These interfaces should cater to how applications typically
interact with other applications and network services rather than
forcing them to use older mechanisms that are more complex to
understand and implement, as well as operate.</t>

<t>The most critical component of the IRS is developing standard data
models with their associated semantics.  While many routing protocols
are standardized, associated data models for IRS are not yet
available.  Instead, each router uses different information,
mechanisms, and CLI which makes a standard interface for use by
applications extremely cumbersome to develop and maintain.  Well-known
data modeling languages, such as YANG <xref target="RFC6020"/>, exist
and might be used for defining the necessary data models; more
investigation into alternatives is required.  It is understood that
some portion (hopefully a small subset) will remain as proprietary
extensions; the data models must support future extensions and
proprietary extensions.</t>

<t>Since the IRS will need to support remote access between
applications running on a host or server and routers in the network,
at least one standard mechanism must be identified and defined to
provide the transfer syntax, as defined by a protocol, used to
communicate between the application and the routing system.  Common
functionality that IRS needs to support includes acknowledgements,
dependencies, request-reserve-commit.</t>

<t>Appropriate candidate protocols must be identified that reduce the
effort required by applications and, preferably, are familiar to
application developers.  Ideally, this should not require that
applications understand and implement existing routing protocols to
interact with IRS.  These interfaces should instead be based on
light-weight, rapidly deployable approaches; technology approaches
must be evaluated but examples could include ReSTful web services,
JSON, XMPP, and XML.  These interfaces should possess self-describing
attributes (e.g. a web services interface) so that applications can
quickly query and learn about the active capabilities of a device.
</t>

<t>It may be desirable to also define the local syntax
(e.g. programming language APIs) that applications running local to a
router can use.</t>

<t>Since evolution is anticipated in IRS over time, it is important
that versioning and backwards compatibility are basic supported
functionality.  Similarly, common consistent error-handling and
acknowledgement mechanisms are required that do not severely limit the
scalability and responsiveness of these interfaces.</t>

</section>

<section title="Common Interface Considerations">

<section title="Capabilities">

<t>Capability negotiation is a critical requirement because different
implementations and software versions will have different abilities.
Similarly, applications may have different capabilities for receiving
exported information.</t>

<t>The IRS will have multiple interfaces, each with their own set of
capabilities.  Such capabilities may include the particular data model
and what operations can be performed at what scale.</t>

<t>The capabilities negotiated may be filtered based upon different
information, such as the application's authorization, application's
capabilities, and the desired granularity for abstraction which the
application understands.  Different types of authorization may require
the router to advertise different capabilities and restrictions.</t>

<t>The capability negotiation may take place at different levels of
detail based upon the application and the specific functions in the
IRS that the application is negotiating.  The router and application
must use the IRS to agree upon the proper level of abstraction for the
interaction. For example, when an application describes a route
between two topological items, these items may vary in detail from a
network domain's name at a high level, or down to the port forwarding
specifics of a particular device.</t>

<t>The data-model and capabilities available for an element may depend
upon whether the element is physical or virtual; the virtual/physical
distinction does not matter to IRS.  Similarly, the location of the
element may influence how an application converses with the associated
router.</t>

</section>

<section anchor="sec_needs" 
title="Identity, Authorization, Authentication, and Security">

<t>Applications that wish to manipulate or interrogate the state of
the routing system must be appropriately authorized.  This means that
at least one means of determining the unique identity of an
application and its associated access privileges must be available;
this implies that the identity and associated access privileges must
be verifiable from the router being programmed.</t>

<t>Furthermore, being able to associate a state and the modifications
to a state with a specific application would aid in troubleshooting
and auditing of the routing system.  By associating identity and
authorization with installed state, other applications with
appropriate authority can clean up state abandoned by failed
applications, if necessary.</t>

<t>Security of communication between the application and the router is
also critical and must be considered in the design of the mechanisms
to support these programmatic interfaces.</t>

</section>

<section title="Speed and Frequency of State Installation">

<t>A programmatic interface does not by itself imply the frequency of
state updates nor the speed at which the state installation is
required. These are critical aspects of an interface and govern what
an application can use the interface for.  The difference between
sub-second responsiveness to millions of updates and a day delay per
update is, obviously, drastic.  The key attributes of the programmatic
interface are described in <xref target="Intro"/> and include that the
interface must be asynchronous.</t>

<t>For each interface in IRS, it will be necessary to specify expected
scaling, responsiveness, and performance so that applications can
understand the uses to which the IRS can be used.</t>

<t>IRS must support asynchronous streaming real-time interactions
between the applications and router.  IRS must assume that there are
many unrelated applications that may be simultaneously using IRS.
This implies that applications must be able to subscribe to change
events that notify them about changes done to state by other
applications or configuration.</t>

<t>Furthermore, IRS should construct interfaces that cater to
different scaling and frequency of update parameters. For example,
slow, but detailed queries of the system, or fast yet higher level
(less detailed) queries or modifications.</t>

</section>

<section title="Lifetime of IRS-Installed Routing System State">

<t>In routers today, the lifetime of different routing state depends
upon how that state was learned and committed.  If the state is
configuration state, then it is ephemeral when just in the running
configuration or persistent when written to the startup configuration.
If the state is learned via a routing protocol or SNMP, it is
ephemeral, lasting only until the router reboots or the state is
withdrawn.</t>

<t>Unlike previous injection mechanisms that implied the state
lifetime, IRS requires that multiple models be supported for the
lifetime of state it installs.  This is because the lifetime or
persistence of state of the routing system can vary based on the
application that programmed it, policies or security authorization of
the application.</t>

<t>There are four basic models to be supported.</t>

<t><list style="hanging">

<t hangText="Ephemeral: ">State installed by the application remains
on the router in its active memory until such time as it is either
removed by a routing or signaling protocol, removed by a configuration
initiated by an application, or the router reboots.  In the case of
the latter, past state is forgotten when the router reboots. </t>

<t hangText="Persistent: ">State installed by the application remains
on the router across reboots or restarts of the system.  It can be
dynamically removed or manipulated by an application, by
configuration, or by the routing system itself.  This state does not
appear in the router's configuration; it is processed after all the
configuration upon a reboot.</t>

<t hangText="Time-Based: ">When state is installed by the application,
it has an expiration time specified.  When that time has passed, the
state is removed from the router.  It can also be dynamically removed
or manipulated by an application, by configuration or the routing
system itself. State that hasn't expired will remain on a router
through reboots.</t>

<t hangText="Time-Based Ephemeral: ">When state is installed by the
application, it has an expiration time specified.  When that time has
passed, the state is removed from the router.  It can also be
dynamically removed or manipulated by an application, by
configuration, by the routing system itself, or by the router
rebooting.  Past state is forgotten after the router reboots.</t>

</list></t>

</section>

<section title="Start-Time of IRS-Installed Routing System State">

<t>To provide flexibility, pre-programming, and handle dependencies,
it is necessary to have multiple models of when a operation is to be
handled.  There are the following basic models to be supported.</t>

<t><list style="hanging">

<t hangText="Immediate: ">When the operation is received, it should
be acted upon as quickly as reasonable (e.g. queued with other
outstanding requests if necessary).</t>

<t hangText="Time-Based: ">An application may provide an operation
that is to be initiated at a particular time.  When the specified time
is reached, the operation should be acted upon as quickly as
reasonable.  Implementations may, of course, strive to improve the
time-accuracy at which the operation is initiated.</t>

<t hangText="Triggered: ">The operation should be initiated when the
specified triggering event has happened.  A triggering event could be
the successful or failed completion of another operation.  A
triggering event could be a system event, such as an interface up or
down, or another event such as a particular route changing its
next-hops.</t>
</list></t>

<t>Because it is possible to request operations in models other than
"Immediate" and some of the start-times will be at an unknown future
point (e.g. "Triggered"), it is not feasible to guarantee that the
resources required by an operation will always be available without
reserving them from the time the operation is received.  While that
type of resource reservation should be possible, applications must
also be able to handle an operation failing or being preempted due to
resources or due to a higher priority or better authorized application
taking ownership of the associated state or resource.</t>

</section>

</section>

<section title="Bidirectional Interfaces to the Routing System">

<t>IRS is a bidirectional programmatic interface that allows both
routing and non-routing applications to install, remove, read, and
otherwise manipulate the state of the routing system.</t>

<t>Just as the Internet routing system is not a single protocol or
implementation layer, neither does it make sense for the IRS to be at
a single layer or reside within a single protocol.  For each protocol
or layer, there are different data models, abstractions and interface
syntaxes and semantics required.  Howeve,r with this in mind, it is
ideal that a minimal set of mechanism(s) to define, transfer and
manipulate this state will be specified with as few optional
characteristics as possible. This will foster better interoperability
between different vendor implementations.</t>

<t>Since IRS is focused on the routing system, the layers of interest
start with the RIB and continue up through the IGPs, BGP, RSVP-TE,
LDP, etc.  The intent is neither to provide interfaces to the
forwarding plane nor to provide interfaces to application layers.</t>

<t>It is critical that these interfaces provide the ability to learn
state, filtered by request, as well as install state.  IRS assumes
that there will be multiple applications using IRS and therefore the
ability to read state is necessary to fully know the router's state.
In general, if an interface allows the setting of state, the ability
to read and modify that state is also necessary.</t>

<section title="Static Routing">

<t>The ability to specify static routes exists via CLI and MIBs but
these mechanisms do not provide a streaming programmatic interface.
IRS solves this problem by proposing interfaces to the RIB, LFIB, and
Multicast RIBs.</t>

<t>By installing static routes into the RIB layer, IRS is able to
utilize the existing router OS and its mechanisms for distributing the
selected routes into the FIB and LIB.  This avoids the need to
model or standardize the forwarding plane.</t>

<section title="Routing Information Base Interface">

<t>The RIB is populated with routes and next-hops as supplied by
configuration, management, or routing protocols.  A route has a
preference based upon the specific source from which the route was
derived.  Static routes, specified via CLI, can be installed with an
appropriate preference.  The FIB is populated by selecting from the
RIB based on policy and tie-breaking criteria.</t>

<t>The IRS interface should allow dynamic reading and writing of
routes into the RIB.  There are several important attributes
associated with doing so, as follows:</t>

<t><list style="hanging">

<t hangText="Preference Value: ">This allows decisions between
conflicting routes, whether IRS-installed or otherwise.  IRS-installed
routes can each be installed with a different preference value.</t>

<t hangText="Route Table Context: ">There can be different route table
contexts in the RIB.  Examples include multiple protocols (e.g. IPv4,
IPv6), multiple topologies, different uses, and multiple networks
(e.g. VRF tables for VPNs).  Appropriate application-level
abstractions are required to describe the desired route table
context.</t>

<t hangText="Route or Traffic Identification">The specific IP prefix
or even interface must be specified.</t>

<t hangText="Outgoing Path and Encapsulation: ">It is necessary to
specify the outgoing path and associated encapsulation.  This may be
done directly or indirectly.  This is one of the more complex aspects
with the following considerations.

<list style="hanging">

<t hangText="Primary Next-Hops: ">To support multi-path forwarding,
multiple primary next-hops can be specified and the traffic flows
split among them.</t>

<t hangText="Indirection: ">Instead of specifying particular primary
next-hops, it is critical to be able to provide the ability for
indirection, such as is used between BGP routes and IGP routes.  Thus,
the outgoing path might be specified via indirection to be the same as
another route's.</t>

<t hangText="Encapsulation: ">Associated with each primary next-hop can
be details on the type of encapsulation for the packet.  Such
encapsulation could be MPLS, GRE, etc. as supported by the router.</t>

<t hangText="Protection: ">For fast-reroute protection, each primary
next-hop may have one or more alternate next-hops specified.  Those
are to be used when the primary next-hop fails.</t>

<t hangText="DSCP: "> For QoS, the desired DSCP to be used for the
outgoing traffic can be specified. </t>

</list></t>

</list></t>

<t>It is useful for an application to be able to read out the RIB
state associated with particular traffic and be able to learn both the
preferred route and its source as well as other candidates with lower
preference.</t>

<t>Although there is no standardized model or specification of a RIB,
it may be possible to build an interoperable bi-directional interface
without one.</t>

</section>

<section title="Label Forwarding Information Base Interface">

<t>The LFIB has a similar role to the RIB for MPLS labeled packets.
Each entry has slightly different information to accommodate MPLS
forwarding and semantics.  Although static MPLS can be used to
configure specific state into the LFIB, there is no bidirectional
programmatic interface to program, modify, or read the associated
state.</t>

<t>Each entry in the LFIB requires a MPLS label context
(e.g. platform, per-interface, or other context), incoming label,
label operation, and next-hops with associated encapsulation, label
operation, and so on.  Via the IRS LFIB interface, an application
could supply the information for an entry using either a pre-allocated
MPLS label or a newly allocated MPLS label that is returned to the
application.</t>

</section>

<section title="Multicast Routing Information Base Interface">

<t>There is no bidirectional programmatic interface to add, modify,
remove or read state from the multicast RIB.  This IRS interface would
add those capabilities.</t>

<t>Multicast forwarding state can be set up by a variety of protocols.
As with the unicast RIB, an application may wish to install a new
route for multicast.  The state to add might be the full multicast
route information - including the incoming interface, the particular
multicast traffic (e.g. (source, group) or MPLS label), and the
outgoing interfaces and associated encapsulations to replicate the
traffic too.</t>

<t>The multicast state added need not match to well-known protocol
installed state.  For instance, traffic received on an specified set,
or all, interfaces that is destined to a particular prefix from all
sources or a particular prefix could be subject to the specified
replication.</t>

</section>

</section>

<section title="Beyond Destination-based Routing">

<t>Routing decisions and traffic treatment is not merely expressable
via destination-based routing or even (S, G) routing, such as in
multicast.  Capturing these aspects into appropriate interfaces for
the IRS provides the ability for applications to control them as
well.</t>

<section title="Policy-Based Routing Interface">

<t>A common feature of routers is the ability to specify policy-based
routing (PBR) rules for accepting, dropping, or differently forwarding
particular traffic.  This is a very useful functionality for an
application to be able to rapidly add and remove state into.  Such
state would indicate the particular traffic to be affected and its
subsequent behavior (e.g. drop, accept, forward on specified outgoing
path and encapsulation, QoS, DSCP marking, policing, etc.).  Such
state is made more complex by the potential importance of ordering
among the PBR rules.</t>

<t>While PBR rules can be specified via CLI, this mechanism is not a
streaming programmatic interface nor is there generally the ability to
specify particular time-based lifetimes for each rule.</t>

</section>

<!-- 
 [Alia 18-July-2012]

 Bruno's suggestion is to also discuss stateful services (which look
 at the entire stream of traffic for a given flow), such as stateful
 firewall, NAT, load balancers, etc. and add APIs.  I can see the
 utility but am not sure of either how we'd model that generically nor
 how to connect it cleanly into routing.  We should discuss.

-->

<section title="QoS State">

<t>While per-hop behaviors are defined as well as standard DSCP
meanings, the details of QoS configuration are not standardized and
can be highly variable depending upon platform.  It is NOT a goal of
this work to standardize QoS configurations.  Instead, a data object
model can define push/pull configurations.  More investigation is
needed to better describe the details.</t>

</section>

</section>

<section title="Protocol Interactions">

<t>Providing IRS interfaces to the various routing protocols allows
applications to specify policy, local topology changes, and
availability to influence the routing protocols in a way that the
detailed addition or modification of routes in the RIB does not.</t>

<t>The decision to distribute the routing state via a routing or
signaling protocol depends upon the protocol-layer at which this state
is injected into the routing system.  It may also depend upon which
routing domain or domains this information is injected as well.</t>

<t>In addition it is necessary to have the ability to pull state
regarding various protocols from the router, a mechanism to register
for asynchronous events, and the means to obtain those asynchronous
events.  An example of such state might be peer up/down.</t>

<section title="IGP Interfaces">

<t>The lack of a streaming programmatic interface to the IGPs limits
the ability of applications to influence and modify the desired
behavior of the IGP.</t>

<t>An application may need to indicate that a router is overloaded
(via ISIS or the method described in <xref target="RFC3137"/>) because
that router does not yet have sufficient state synchronized or
installed into it.  When critical state is provided not merely by
routers but also from applications via the IRS, a synchronization
mechanism can be needed.</t>

<t>The ability for an application to modify the local topology can be
part of this interface.  One possibility is to allow modification of
local interface metrics to generally influence selected routes.  A
more extensive interface might include the ability to create a OSPF or
ISIS adjacency across a specified interface (virtual or real) with the
appropriate associated encapsulation.</t>

<t>The ability to attach a prefix to the local router would provide a
straightforward method for an application to program a single router
and have the proper routes computed and installed by all other routers
in the relevant domains.  Additional aspects to the prefix attachment,
such as the metric with which to attach the prefix and fast-reroute
characteristics, would be part of the interface.</t>

<t>Beyond such pure routing information, the need for an application
to be able to install state to be flooded via an IGP has already been
recognized.  <xref target="I-D.ietf-isis-genapp"/> specifies a
mechanism for flooding generalized application information via ISIS,
but does not describe how an application can generate or consume this
information.  Similarly, <xref target="RFC5250"/> specifies Opaque
LSAs for OSPF to provide for application-specific information to be
flooded.  An IRS interface and associated data object model would
provide such a mechanism.</t>

<t>Additional investigation will identify other state that
applications may wish to install.</t>

<t>From the IGP, applications via IRS can extract significant
topological information about the routers, links, and associated
attributes.</t>

</section>

<section title="BGP Interface">

<t>BGP carries significant policy and per-application specific
information as well as internet routes.  A significant interface into
BGP is expected, with different data object models for different
applications.  For example, the IRS interface to BGP could provide the
ability to specify the policy on which paths BGP chooses to advertise.
Additionally, the ability to specify information with an
application-specified AFI/SAFI could provide substantial flexibility
and control.</t>

<t>An existing example of application information carried in BGP is
BGP Flowspec <xref target="RFC5575"/> which can be used to provide
traffic filtering and aid in handling denial-of-service attacks.</t>

<t>The ability to extract information from BGP is also quite critical.
A useful example of this is the information available from BGP via
<xref target="I-D.gredler-idr-ls-distribution"/>, which allows
link-state topology information to be carried in BGP.</t>

</section>

<section title="PIM and mLDP Interfaces">

<t>For PIM and mLDP, there are at least two types of state that an
application might wish to install.  First, an application might add an
interface to join a particular multicast group.  Second, an
application might provide an upstream route for traffic to be received
from - rather than having PIM or mLDP need to consult the unicast
RIB.</t>

<t>Additional investigation will identify other state that
applications may wish to install.</t>

</section>

</section>

<section title="Triggered Sessions and Signaling">

<section title="OAM-related Sessions Interface">

<t>An application may need to trigger new OAM sessions (e.g. BFD,
VCCP, etc.) using an appropriate template.  For example, there may be
applications that need to create a new tunnel, verify its
functionality via new triggered OAM sessions, and then bring it into
service if that OAM indicates successful functionality. More
investigation is needed to better describe the details. </t>

</section>

<section title="Dynamic Session Creation">

<t>An application may wish to trigger a peering relationship for a
protocol.  For instance, a targeted LDP session may be required to
exchange state installed locally with a remote router.  More
investigation is needed to better describe the different cases and
details. </t>

</section>

<section title="Triggered Signaling">

<t>To easily create dynamic state throughout the network, an
application may need to trigger signaling via protocols such as
RSVP-TE.  An example of such an application can be a Stateful Path
Computation Element (PCE)<xref target="I-D.ietf-pce-stateful-pce"/>,
which has control of various LSPs that need to be signaled.</t>

<t>More investigation is needed to better describe the different cases
and details. </t>

</section>

</section>

</section>


<section 
title="Interfaces for Learned Information from the Routing System">

<t>Just as applications need to inject state into the routing system
to meet various application-specific and policy-based requirements, it
is critical that applications be able to also extract necessary state
from the routing system.</t>

<t>A part of each of these interfaces is the ability to specify the
generation of the desired information (e.g., collecting specific
per-flow measurements) and the ability to specify appropriate filters
to indicate the specifics and abstraction level of the information to
be provided</t>

<t>The types of information to extract can be generally grouped into
the following different categories.</t>

<t><list style="hanging">

<t hangText="Topological: ">The need to understand the network
topology, at a suitable abstraction layer, is critical to
applications.  Connectivity is not sufficient - the associated costs,
bandwidths, latencies, etc. are all important aspects of the network
topology that strongly influence the decision-making and behavior of
applications.</t>

<t hangText="Measurements: ">Applications require measurements of
traffic and network behavior in order to have a more meaningful
feedback control loop.  Such information may be per-interface,
per-flow, per-firewall rule, per-queue, etc.</t>

<t hangText="Events: ">There are a variety of asynchronous events that
an application may require or use as triggering conditions for
starting other operations. An obvious example is interface state
events.</t>

<t hangText="Configuration: ">For some aspects, it may be necessary
for applications to be able to learn about the routing configuration
on a box.  This is partially available via various MIBs and NetConf.
What additional information needs to be exported and the appropriate
mechanisms needs further examination.</t>

</list></t>

<t>The need to extract information from the network is not new; there
is on-going work in the IETF in this area.  This framework describes
those efforts in the context of the above categories and starts the
discussion of the aspects still required.</t>

<section title="Efforts to Obtain Topological Data">

<t>Topological data can be defined and presented at different layers
(e.g. Layer-2, Layer-3) and with different characteristics exposed or
hidden (e.g. physical or virtual, SRLGs, bandwidth, latency, etc.).
It can also have different states, such as configured but unavailable,
configurable, active, broken, administratively disabled, etc.</t>

<t>To solve the problem of only being able to obtain topological data
via listening to the IGP in each area, BGP-LS <xref
target="I-D.gredler-idr-ls-distribution"/> defines extensions to BGP
so that link-state topology information can be carried in BGP and a
single BGP listener in the AS can therefore learn and distribute the
entire AS's current link-state topology.  BGP-LS solves the problem of
distributing topological information throughout the network.  While
IRS may expand the information to be distributed, IRS addresses the
API aspect of BGP-LS and not the network-wide distribution.</t>

<t>At another level, ALTO <xref target="RFC5693"/> provides topological
information at a higher abstraction layer, which can be based upon
network policy, and with application-relevant services located in it.
The mechanism for ALTO obtaining the topology can vary and policy can
apply to what is provided or abstracted.</t>

<t>Neither of these fully meet the need to obtain detailed, layered
topological state that provides more information than the current
functional status.  While there are currently no sufficiently complete
standards, the need for such functionality can be deduced by the
number of proprietary systems that have been developed to obtain and
manage topology; even Element Management Systems start with the need
for learning and manipulating the topology.  Similarly, orchestration
layers for applications start with the need to manage topology and the
associated database.</t>

<t>Detailed topology includes aspects such as physical nodes, physical
links, virtual links, port to interface mapping, etc.  The details
should include the operational and administrative state as well as
relevant parameters ranging from link bandwidth to SRLG membership.
Layering is critical to provide the topology at the level of
abstraction where it can be easily used by the application.</t>

<t>A key aspect of this interface is the ability to easily rate-limit,
filter and specify the desired information to be extracted.  This will
help in allowing the interface to scale when queries are done.</t>

</section>

<section title="Measurements">

<t>IPFIX <xref target="RFC5470"/> provides a way to measure and export
per-traffic flow statistics.  Applications that need to collect
information about particular flows thus have a clear need to be able
to install state to configure IPFIX to measure and export the relevant
flows to the appropriate collectors.</t>

</section>

<section title="Events">

<t>A programmatic interface for application to subscribe to
asynchronous events is necessary.  In addition to the interface state
events already mentioned, an application may wish to subscribe to
certain OAM-triggered events that aren't otherwise exported. </t>

<t>A RIB-based event could be reporting when the next-hops associated
with a route have changed.  Other events could be used to verify that
forwarding state has been programmed.  For example, an application
could request an event whenever a particular route in the RIB has its
forwarding plane installation completed.</t>

<t>When an application registers for events, the application may
request to get only the first such event, all such events, or all
events until a certain time.</t>

<t>The full set of such events, that are not specifically related to
other interfaces, needs to be investigated and defined.</t>

</section>

</section>

<section title="Manageability Considerations">

<t>Manageability plays a key aspect in IRS.  Some initial examples
include:

<list style="hanging">

<t hangText="Data Authorization Levels: ">The data-models used for IRS
need the ability to indicate the required authorization level for
installing or reading a particular subset of data.  This allows
control of what interactions each application can have.</t>

<t hangText="Identity Authorization Levels: ">Associated with an
application's identity should be an identity authorization level that
is in a heirarchy so that higher authorized applications can manage
and remove the state and resources used by other applications.  The
top of such a heirarchy would be the router configuration itself.</t>

<t hangText="Resource Limitations: ">Using IRS, applications can
consume resources, whether those be operations in a time-frame,
entries in the RIB, stored operations to be triggered, etc.  The
ability to set resource limits based upon authorization is critical.</t>

<t hangText="Configuration Interactions: ">The interaction of state
installed via the IRS and via a router's configuration needs to be
clearly defined.</t>
</list></t>

</section>

    <section anchor="IANA" title="IANA Considerations">
      <t>This document includes no request to IANA.</t>
    </section>

    <section anchor="Security" title="Security Considerations">

      <t>This framework describes interfaces that clearly require
      serious consideration of security.  The ability to identify,
      authenticate and authorize applications that wish to install
      state is necessary and briefly described in <xref
      target="sec_needs"/>.  Security of communications from the
      applications is also required.</t>

      <t>More specifics on the security requirements requires further
      investigation.</t>

    </section>

<section anchor="Acknowledgements" title="Acknowledgements">
  <t>The authors would like to thank Ken Gray, Adrian Farrel, Bruno
  Rijsman, Rex Fernando, Jan Medved, John Scudder, and Hannes Gredler
  for their suggestions and review.</t>

</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">

    </references>
-->

    <references title="Informative References">
    &RFC3137;
    &RFC5250;
    &RFC5470;
    &RFC5575;
    &RFC5693;
    &RFC6020;
    &I-D.ietf-pce-stateful-pce;
    &I-D.ietf-isis-genapp;
    &I-D.gredler-idr-ls-distribution;
    </references>

    <!-- Change Log

v00 2012-07-11  AKA   Initial version
v0i 2012-07-25  AKA   Adding in Adrian's and Dave Meyer's comments
    -->

  </back>
</rfc>