One document matched: draft-xie-alto-sdn-extension-use-cases-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" [
]>
<?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="no"?>
<!-- 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-xie-alto-sdn-extension-use-cases-00" ipr="trust200902">
  <!-- category values: std, bcp, info, exp, and historic
     ipr values: trust200902, noModificationTrust200902, noDerivativesTrust200902
     pre5378Trust200902
     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="ALTO Use Cases For SDN">Use Cases for ALTO with Software Defined Networks</title>

    <author fullname="Haiyong Xie" initials="H." surname="Xie">
      <organization>Huawei & USTC</organization>
      <address>
        <postal>
          <street>2330 Central Expy</street>

          <!-- Reorder these if your country does things differently -->

          <city>Santa Clara</city>

          <region>CA</region>

          <code>95050</code>

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

        
        <email>Haiyong.xie@huawei.com</email>

        <!-- uri and facsimile elements may also be added -->
      </address>
    </author>
    
    <author fullname="Tina Tsou" initials="T." surname="Tsou">
      <organization>Huawei (USA)</organization>
      <address>
        <postal>
          <street>2330 Central Expy</street>

          <!-- Reorder these if your country does things differently -->

          <city>Santa Clara</city>

          <region>CA</region>

          <code>95050</code>

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

        
        <email>Tina.Tsou.Zouting@huawei.com</email>

        <!-- uri and facsimile elements may also be added -->
      </address>
    </author>
    
    <author fullname="Diego R. Lopez" initials="D.R." surname="Lopez">
      <organization>Telefonica I+D</organization>
      <address>
        <postal>
          <street>Don Ramon de la Cruz, 84</street>

          <!-- Reorder these if your country does things differently -->

          <city>Madrid</city>

          <region></region>

          <code>28006</code>

          <country>Spain</country>
        </postal>

        

        <email>diego@tid.es</email>

        <!-- uri and facsimile elements may also be added -->
      </address>
    </author>
   
    <author fullname="Hongtao Yin" initials="H." surname="Yin">
      <organization>Huawei (USA)</organization>
      <address>
        <postal>
          <street>2330 Central Expy</street>

          <!-- Reorder these if your country does things differently -->

          <city>Santa Clara</city>

          <region>CA</region>

          <code>95050</code>

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

        
        <email>Hongtao.yin@huawei.com</email>

        <!-- uri and facsimile elements may also be added -->
      </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>General</area>

    <workgroup>Internet Engineering Task Force</workgroup>

    <!-- WG name at the upperleft corner of the doc,
         IETF is fine for individual submissions.  
    If this element is not present, the default is "Network Working Group",
         which is used by the RFC Editor as a nod to the history of the IETF. -->

    <keyword>ALTO</keyword>
    <keyword>software defined networks</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>
			
			The introduction of SDN fundamentally changes the way that the ALTO
			works.

			This draft describes the Vertical Architecture and the Horizontal
			Architecture allowing coherent coexistence of application layer
			traffic optimization (ALTO) with software defined network (SDN).  
			
			Unique requirements for design and operations are identified and
			summarized, suggesting that the Vertical Architecture allows better
			division, management, flexibility, privacy control and long-term
			evolution of the network.  
			
			We also define the main interactions and information flows, and
			present a set of use cases to illustrate how we extend ALTO to
			support SDN, in the Vertical Architecture.

		</t>
    </abstract>

  </front>

  <middle>


<section anchor="intro" title="Introduction">

   <t>The concept of Software Defined Network (SDN) has emerged and become 
   one of the fundamental, promising networking primitives that allow 
   flexibility of control, separation of functional planes and continuous 
   evolution in networks. </t>
   
   <t>One of the key features of SDN is the full separation of two functional
   planes: the control plane and the data-forwarding plane. SDN requires 
   that networking devices support such separation, implementing the data 
   plane mechanisms, while the control plane is provided by an external
   entity called the "controller". The other key feature of SDN is the new 
   interaction process between the separated control plane and data-forwarding
   plane. This interaction is mandated to be performed specific open protocols,
   allowing for a free combination of networking devices and controllers, as 
   well as supporting the controller to take decisions on information additional 
   to the networking device status.</t>

   <t>There could be numerous benefits as a result of the above features in
	   SDN, e.g., just to name a few, network virtualization, better flexible
	   control and utilization of the network, networks customized for
	   scenarios with specific requirements. For instance, some SDN
	   technologies have started to find their ways into Data Center Networks
	   (DCNs).  Furthermore, in order to allow efficient and flexible
	   implementation of the above separation and interactions, a significant
	   portion of the SDN system could typically be implemented in software, as
	   opposed to the hardware-based implementation adopted by most of today's
	   networking devices.</t> 

   <t>Due to the great potentials of SDN and the unique requirements of DCNs,
	   Data Centers are likely to become a first place where SDN could be
	   deployed.  We envision that SDN could be gradually adopted by enterprise
	   networks and then by carrier networks due to its unique, attractive
	   features. When deploying SDN in large-scale distributed networks, we
	   expect to see a collection of deployments limited in relatively small
	   segments of a bigger network. In other words, it is likely that the
	   operator of a large-scale enterprise / carrier network prefers to divide
	   the whole networks into multiple smaller segments and put each of there
	   segments in an SDN domain. These smaller network segments (therefore
	   their corresponding SDN domains) are connected and jointly form the
	   larger whole network. Such a divide-and-conquer methodology not only
	   allows gradual deployment and continuous evolution, but also enables
	   flexible provisioning of the network.</t>

   <t>With the deployment of SDN, application layer traffic optimization (ALTO) 
   faces new challenges including, but not limited to, privacy preservation, 
   granularity of information collection and exchange, join optimization, etc. 
   We shall elaborate these challenges and their impacts on the design of ALTO 
   extensions for SDN in this draft.</t>
 
   <section anchor="term" title="Terminology">

	  <t>While the definition of software defined networks is still "nebulous"
		  to some extent, we refer to as SDN the networks that implement the
		  separation of the control and data-forwarding planes and software
		  defined interactions between these two separated planes; such
		  interactions are characterized by open interfaces which allow
		  programming the network equipment's forwarding plane by external
		  agents, e.g., SDN controllers.  However, how the two separated planes
		  interact is not a focus of this draft; instead, the ALTO extension
		  for SDN recommended in this draft is independent of how such
		  interactions would be.</t>

	  <t>An SDN domain is a portion of a network infrastructure, consisting of
		  numerous connected networking devices that are SDN capable (i.e., SDN
		  capable devices implement the control/forwarding plane separation and
		  the open interfaces allowing external agents to program the devices)
		  and a domain controller which implements the SDN control plane
		  functionalities for these devices. An SDN domain can be as small as a
		  sub-network of a dozen devices or as large as a medium/large data
		  center network.</t>

	  <t>A software defined network is a network that has one or multiple SDN
		  domains. Due to an SDN domain typically has limited coverage, an SDN
		  may be comprised of networking devices under control of some SDN
		  domains (i.e., SDN managed devices) and devices under no control of
		  any SDN domain (i.e., normal devices). Note that such normal devices
		  could still be SDN capable and their control/forwarding planes are
		  managed as in normal networks today. This implies that a network with
		  both normal devices and SDN capable devices (managed by SDN domains)
		  needs both normal and SDN capable control/forwarding plane
		  management.</t>

   </section>  <!-- term -->
</section>  <!-- intro -->


<section anchor="overview" title="An Overview of Software Defined Network">

   <t>This section provides a high level and conceptual overview of 
   software defined network in order to help illustrate its unique 
   requirements for ALTO.</t>

   <section anchor="descrip" title="Software Defined Network">

      <t>We refer to as an "SDN" a carrier's or an enterprise's network which
      deploys or implements software defined networking technologies.</t> 

      <t>Since SDN separates the control plane and data-forwarding plane, we
      refer to the entity that implements the control-plane functionalities
      as the "SDN controller".</t>   

      <t>In order for a network to be classified as an SDN, it is unnecessary 
      that all networking devices have to be SDN capable. Some of devices
      in a network may be managed by an SDN controller while the remaining 
      ones may not; such a network is still qualified as an SDN.</t>

	  <t>Applications in software defined networks are either SDN-aware or
		  unaware of SDN.  
		  
		  <list style="symbols">   

			  <t>If an application is SDN-aware, then the application may
				  prefer direct communication with SDN controllers, which
				  implies that there must exist mechanism(s) for SDN-aware
				  applications to locate and communication with SDN
				  controllers. If the application prefers indirect
				  communication with SDN controllers, then it follows the case
				  of SDN-unaware applications (see below). 
				  
				  Applications that are SDN-aware may be able to better utilize
				  the SDN capable network due to its knowledge about SDN and
				  its capability of proactive, direct interaction with SDN.</t> 
			  
			  <t>If an application is SDN-unaware, then the application
				  indirectly communicates with SDN controllers by sending
				  application datagrams with specific formats, via which the
				  application can specify its requirements for the network
				  resources. 
				  
				  Legacy applications (applications for the current
				  IP networks) are largely SDN-unaware, and such applications
				  may not be able to utilize the SDN capable networks as
				  efficiently as SDN-aware applications.</t> 
		  
		  </list>    
      Whether and how applications should/can be SDN-aware or SDN-unaware is
      beyond the scope of this draft. However, the framework we described in 
      this draft is applicable to both SDN-aware and SDN-unaware cases.</t>

   </section>  <!-- descrip -->

   <section anchor="divis" title="Division of Network">

      <t>A network operator may decide to divide the network into multiple 
      sub-networks, some of which are SDN capable and thus are managed by
      corresponding SDN controllers.</t> 

      <t>There could be numerous reasons for such division of network. Below we 
      summarize a few of them:
      <list style="symbols">
        
          <t>Scalability.
          <vspace blankLines="1"/>
          The number of devices an SDN controller can feasibly manage is likely
          to be limited. Therefore, in order to manage a many devices that cannot
          be put under control of a single SDN controller, multiple controllers
          have to be deployed. Hence, the network is divided into multiple 
          sub-networks; if a sub-network has SDN capable devices, it should
          be managed by an SDN controller.</t>
  
         <t>Manageability.
         <vspace blankLines="1"/>
         At the network level, in order to reduce the complexity of management,
         a carrier may decide to divide the network into multiple sub-networks
         and put some of them under control of some SDN controllers (provided
         that the devices in such sub-networks are SDN capable); each of
         the sub-networks can be managed independently to some extent, thus 
         reducing the overall complexity of managing the whole network. Even
         at the sub-network level, a carrier may still decide to further divide
         the sub-network in order to further reduce complexity of management.
         For instance, a sub-network under control of an SDN controller may
         span across a large geographical area or cover a large number of 
         devices; in this case it is reasonable for the carrier to further
         divide it into two or even more sub-networks, such that the 
         complexity of managing each individual sub-network plus the overall
         complexity of managing all divided sub-networks are reduced.</t>
  
         <t>Privacy.
         <vspace blankLines="1"/>
         When a carrier divides its network into multiple sub-networks and put
         them under control of SDN, the carrier may choose to implement 
         difference privacy policies in different sub-networks. For example, 
         a carrier may dedicate a part of its infrastructure to some certain 
         customers, dividing the whole network and dedicate some of the 
         sub-networks is a convenient and scalable way to manage the network 
         resources for these customers. Another example is that a sub-network
         in a carrier's network may be dedicated to some certain customers 
         and such information as network topology may not be disclosed to any
         external entity.</t>
  
         <t>Deployment
         <vspace blankLines="1"/>
         The deployment of network infrastructures, especially new network
         infrastructure and technologies, has to be incremental. This means 
         that at any time, a carrier's network may consist of a portion of 
         legacy and a portion of non-legacy infrastructure. When deploying new
         infrastructure or technologies, it is highly preferable to limit the 
         scope of new deployment and do it in an incremental way. In such cases,
         it is very favorable to divide the network into multiple individually
         manageable sub-networks and choose some of them to deploy the new
         infrastructure / technologies.</t>
        
      </list>
      </t>

   </section>  <!-- divis -->

   <section anchor="domain" title="SDN Domain">

      <t>With the introduction of SDN, we believe that it is inevitable for 
      carriers to divide their networks for many reasons as illustrated in the
      preceding subsection. Therefore, we believe that to better suit this need,
      it should be recommended that SDN domains are defined and applied in 
      division of the networks.</t>

      <t>An SDN domain is a sub-network, resulted from division of a network 
      which is determined by the network operator. Each domain typically 
      consists of numerous connected networking devices, each of which is SDN
      capable. Each domain also has a domain controller which implements SDN
      control plane functionalities for the devices in this domain. Another
      important task such a domain controller is responsible for includes 
      fine-grain network information collection (for devices in this domain).
      The collected information is necessary for the controller to make 
      data-forwarding plane decisions. Note that such a domain controller may
      be integrated as a part of a so-called "network operating system" (NOS).
      An example of such a network operating system is illustrated in 
      <xref target="onix"/>.</t>

      <t>Any networking device, if under the control of certain SDN domains,
      should belong to only one SDN domain and should be under the control of
      the SDN domain's controller. In other words, the intersection of two
      domains is always empty.</t> 

	  <t>Furthermore, SDN domains are connected (via the connectivity among
		  underlying devices provided by the underlying network; such devices
		  belong to different SDN domains) to form the whole network. For a
		  large-scale distributed networks owned by a national/multi-national
		  carrier or enterprise, it is natural to adopt the
		  "divide-and-conquer" strategy and divide the whole network into
		  multiple SDN domains. Even for small or medium networks, multiple SDN
		  domains may be necessary in order to virtualize the network resources
		  (e.g., set up multiple SDN domains for a large data center network to
		  instantiate multiple virtual data centers for many content
		  providers).  Note that how multiple SDN domains inside a
		  carrier's/enterprise' network work together is beyond the scope of
		  this draft and is handled by other working groups.</t>

      <t>Inside each SDN domain, its controller may define domain-specific 
      policies on information importing from devices, aggregating, and exporting
      to external entities. Such policies may not be made public; therefore, 
      other domain controllers or ALTO may not know the existence of such
      policies for any given SDN domain.</t>

      <t>The natural network division implemented by SDN domains impose 
      significantly new and challenging requirement for shaping the interactions
      between SDN and ALTO, and therefore designing the protocols to enable 
      such interactions.</t>

   </section>  <!-- domain -->
</section>  <!-- overview -->

<section anchor="architecture" title="Architectural Considerations for SDN and ALTO">

	<t>We introduce two architectures that allow coherent co-existence of SDN
		and ALTO in this section:

		<list style="symbols">

			<t> The Vertical Architecture (or the V Architecture for short)
				allows better division, management, flexibility, privacy
				control and long-term evolution of the network, and </t>

			<t> The Horizontal Architecture (or the H Architecture for short)
				simplifies the implementation of ALTO extensions for SDN. </t>

		</list>

		We next present each of these two architectures individually.  
	</t> 

	<section anchor="v-arch" title="The Vertical Architecture"> 

		<t> 
			
			The Vertical Architecture is illustrated in the following figure.
			The network has one or multiple SDN domains and an ALTO server. The
			interactions between the SDN controllers and the SDN capable
			devices fall in the scope of SDN and therefore is out of the scope
			of this draft; instead, the interactions between the SDN domains
			(more specifically, SDN controllers) and the ALTO server is what
			this draft focuses on.  
		
		</t>

		<figure src='v-arch.png' alt='[The Vertical Architecture]'>
			<!--preamble>This is the preamble.</preamble-->
			<artwork>
.----------------------------------------------.
|                ALTO Server                   |
`----------------------------------------------'
              ^            |
.-------------|------------|-------------------.
|             |            V                   |
|     .-------------------------------.        |
|     |          SDN Controller       |        |
|     `-------------------------------'        |
|                    |                         |
|                    |                         |
|     .-------------------------------.        |
|     |       SDN Capable Devices     |        |
|     `-------------------------------'        |
|                                              |
|                        An SDN Domain         |
`----------------------------------------------'
			</artwork>
			<!--postamble>This is the postamble.</postamble-->
		</figure> 
		
	  <t>The Vertical Architecture is a hierarchical architecture consisting of
		  three tiers.  In the first tier, the only entity is the ALTO server.
		  In the second tier, the only entities are the SDN domain controllers.
		  In the third tier, the only entities are SDN domains (each domain
		  consists of numerous networking devices).  
		  
		  In this architecture, all entities are owned by the same carrier.
		  However, some of the SDN domains (and the networking devices in them)
		  may be dedicated to certain customers of the carrier (the carrier
		  gives the customers privileges access).  This is subject to a
		  collaboration agreement between them.</t>

		<t> 

			The interactions between the SDN controllers and the ALTO server
			can be divided into two categories: 
			
			<list style="symbols">

				<t> 

					Upward interactions (i.e., from SDN controllers to ALTO
					servers): each SDN controller collects network information
					from the devices managed by it and information from other
					SDN controllers), and report such information to the ALTO
					server, subject to the information aggregation and privacy
					policies defined for the corresponding individual SDN
					domain. Such network information is referred to the
					inter-domain network information. The network information
					could include key information such as domain-level network
					cost, bandwidth, domain-specific connectivity, etc. The
					upward interactions could be implemented in either the push
					model or the pull model. 
					
					It is worth noting that network information collection has
					not been explored, and that network information collection
					could introduce significant overhead and complexity, in the
					current scope of ALTO.  However, automated network
					information collection is a key to the success of ALTO.

					With the help of SDN and the Vertical Architecture, such
					automated network information collection becomes feasible
					and appealing. Note that this does not exclude the
					possibility of network operators manually or automatically
					update the ALTO server with the network information (e.g.,
					the network cost map).

					It is also worth noting that an SDN controller may choose
					to report its domain-specific network information only
					(referred to as the intra-domain information), with or
					without privacy policies. In this case, SDN controllers
					become an automated information collector for the ALTO
					server.
				
				</t>

				<t> 
					
					Downward interactions (i.e., from ALTO servers to SDN
					controllers): each SDN controller is also an ALTO client
					and retrieves relevant network information from the ALTO
					server. This is similar to the current scope of ALTO
					without the existence of SDN; however, the differences are
					that with the existence of SDN, the network information is
					generally specific to SDN and SDN domains; SDN controllers
					as ALTO clients could query the ALTO server for either
					inter-domain or intra-domain network information (provided
					that intra-domain information is reported and made
					available).

				</t>

			</list>
		</t>

		<t> 
			
			We refer to as the "upward flow" the information flow from the
			second tier (SDN controllers as ALTO clients) to the first tier
			(ALTO server), and refer to as the "downward flow" the information
			flow in the reverse direction, i.e., from the first tier (ALTO
			server) to the second tier (SDN controllers as ALTO clients).  
		
		</t>

   </section>  <!-- v-arch -->

   <section anchor="h-arch" title="The Horizontal Architecture">

		<t> 
			
			The Horizontal Architecture is illustrated in the following figure.
			Each SDN controller is integrated with an ALTO server. The
			interactions between the SDN controllers and the ALTO server is
			represented by the horizontal line in the figure. An example of
			this architecture can be found in <xref target="abstr"/>.

		</t>

		<figure src='h-arch.png' alt='[The Horizontal Architecture]'>
			<!--preamble>This is the preamble.</preamble-->
		<artwork>

.---------------------------------------------------------------------.
|   .--------------------------.       .--------------------------.   |
|   |     SDN Controller       |〈----|        ALTO Server         |   |
|   `--------------------------'       `--------------------------'   |
`-----------------|---------------------------------------------------'
                  |               
    .-------------------------------.
    |       SDN Capable Devices     |
    `-------------------------------'
			</artwork>
			<!--postamble>This is the postamble.</postamble-->
		</figure>

		<t> 
			
			In the Horizontal Architecture, the SDN controller can act as an
			ALTO client and query the network information of the networking
			devices from the ALTO server. 
			
			However, such network information may not meet the SDN controllers'
			granularity requirement (i.e., the information provided by the ALTO
			server may not be as fine-grained as needed by the SDN controllers). 

			In addition, there may exist redundant information collection, as
			SDN controllers are inevitably collecting various fine-grain
			information from the devices they manage; the information
			collection by the ALTO server, either manually or automatically, is
			likely to be redundant.

			Furthermore, when the carrier decides to divide its network into
			multiple SDN domains, it can be difficult, if not impossible at
			all, for the Horizontal Architecture to adapt to the network
			division.
		
		</t>

   </section>  <!-- h-arch -->

   <section anchor="arch-summary" title="Summary">

	   <t> 
		   
		   The ALTO server covers a carrier's network as a whole; however, if
		   the carrier divide the network into multiple SDN domains, each SDN
		   domain covers only a segment of the network. Additionally, the ALTO
		   server has only relatively coarse-grained information, while SDN
		   domain controllers could easily collect more fine-grained
		   information.  
		   
		   More importantly, different SDN domains may implement different
		   information aggregation and privacy policies (e.g., when such
		   domains are dedicated to certain customers of the carrier). 

	   </t> 
	   
	   <t> 
		   
		   These observations suggest that the Vertical Architecture is
		   advantageous over the Horizontal Architecture. With the Vertical
		   Architecture, SDN and ALTO are explicitly separated and as a result
		   the logic is cleaner and this allows them to evolve independently.
		   Furthermore, the Vertical Architecture makes automated information
		   collect possible for ALTO, which could make ALTO deployment and
		   management easier and more attractive.

		   Therefore, in the remainder of this draft, we mainly focus on the
		   Vertical Architecture. We will describe the interactions and the
		   information flows in further details for the Vertical Architecture.
		
		</t> 
		

   </section>  <!-- arch-summary -->

</section>  <!-- architecture -->


<section anchor="interact" title="Interactions between SDN and ALTO">

	<t> 
		
		We now describe the interactions between SDN and ALTO in details. We
		first compare the ALTO scopes without and with the existence of SDN,
		and then describe the various interactions existing in the Vertical
		Architecture. 
	
	</t>

	<section anchor="interact-scopes" title="ALTO Scopes">

		<t> 
			The existence of SDN differentiates two scopes of ALTO, namely, the
			current scope of ALTO without SDN (referred to as the
			SDN-unfriendly Scope) and the new scope of ALTO with coherent
			coexistence of SDN (referred to as the SDN-friendly Scope): 
			
			<list style="symbols"> 
				
				<t> 
					
					the SDN-unfriendly Scope: in the current scope of ALTO,
					there exist interactions between ALTO clients and ALTO
					servers. Such interactions are one-way interaction, meaning
					that ALTO information flows in one direction, i.e., from
					the server to the clients. More specifically, ALTO clients
					submit ALTO requests to (and subsequently receive ALTO
					responses from) an ALTO server.  
					
					Additionally, ALTO clients in the current scope of ALTO are
					network applications who would like to consume the network
					resources. ALTO clients are typically managed by network
					applications rather than by network carriers.  However,
					ALTO servers are owned and managed by network carriers.  
				
				</t> 

				<t> 
					
					With the introduction of SDN, the interactions between ALTO
					clients and ALTO servers become more complex. A more
					careful examination as well as important ALTO extensions
					are necessary to make ALTO work in the context of SDN.  
					
					It is important to note that if the network does not
					implement network division (i.e., does not implement SDN
					domains), the interactions are "compressed" into a compact
					set of interactions; specifically, both the SDN controller
					(recall that there exists only one single controller for
					the whole network) and the ALTO server could be integrated
					in many equally efficient fashions. For instance, ALTO
					server could be put underneath the controller, i.e., ALTO
					server provides information to the controller, as suggested
					by <xref target="abstr"/>.  
					
					However, if the network implements division via SDN domains
					(i.e., there exists multiple SDN domains), we essentially
					"unfold" the compressed interaction space and need more
					complex structures that allow efficient design and
					implementation, due to the facts that we listed in the
					preceding sections.  Furthermore, the design originally for
					the compressed space could be instantiated for the unfolded
					space but could not be as efficient.
			
				</t> 

			</list>

			We next focus on the SDN-friendly Scope and highlight the complex
			structures and the important differences.
		</t>

	</section> <!-- interact-scopes -->
	
	<section anchor="clients" title="ALTO clients">

         <t>With the existence of SDN and SDN controllers, ALTO clients could 
         be not only legacy network applications (or proxies for these network
         applications), but also SDN controllers.</t> 

         <t>In SDN, SDN controllers have similar needs as the legacy ALTO clients 
         which communicate with ALTO servers, because ALTO clients would like to
         better understand the network and thus improve the application 
         performance.</t> 

         <t>There are multiple reasons for this similarity. The most important 
         reason is that each SDN controller is only responsible for managing a
         sub-network in a carrier's network; therefore, it may not understand 
         well other sub-networks in the same carrier network. However, in order
         to allocate the network resources to satisfy application requirements 
         (note that the end points of such applications may well span across
         multiple SDN domains), an SDN controller may have to involve other
         SDN controllers because the network paths needed may traverse multiple
         SDN domains. Thus, obtaining global information from the ALTO server 
         is a significantly more efficient and more preferable than otherwise
         from SDN interconnection protocols; plus, such protocols do not exist
         yet today.</t>

         <t>Although SDN controllers have similar needs as legacy ALTO 
         applications do, the fundamental properties of SDN controllers as ALTO
         clients are significantly different. One of the differences is the 
         ownership and management, as most SDN controllers require additional
         (and more likely complex) access privileges. Specifically, SDN 
         controllers are typically owned by the network carriers who also own
         their ALTO servers, while legacy ALTO clients are network applications
         typically not owned by network carriers (there are cases where owned
         collaboratively amongst operators).</t> 

         <t>In terms of management, the entity managing SDN controller is the 
         same as that managing the ALTO server. We emphasize that when an SDN 
         domain is dedicated to some customers of a network carrier, the use 
         of the SDN domain is owned by these customers and so is the management.
         In this case, the SDN controllers as ALTO clients are slightly more 
         like legacy ALTO clients. However, there still exist fundamental 
         differences which we will illustrate later.</t>

      </section>  <!-- clients -->

	  <section anchor="SDNiALTO" title="Interactions between SDN and ALTO"> 
		  
		  <t> 
			  
			  Another difference is the interactions between SDN controllers as
			  ALTO clients and ALTO servers. Legacy ALTO clients retrieve
			  information from ALTO servers. However, SDN controllers may also
			  need to push information to ALTO servers. In a carrier's network,
			  SDN controllers and the ALTO server are owned by the same
			  carrier. Interactions between them could be significantly more
			  complex.

		  </t>

		  
		  <section anchor="downward-interaction" title="Downward Interaction"> 
			  <t> 
				  
				  The downward interaction is the information flow from ALTO
				  servers to ALTO clients (i.e., SDN controllers). SDN
				  controllers request information from ALTO servers, similar to
				  legacy ALTO clients. However, the requested information is
				  significantly different.  
				  
				  The fundamental difference is a result of SDN and SDN domain
				  division, which do not exist in legacy network application
				  scenarios. For instance, an SDN controller for a specific SDN
				  domain may be interested in obtaining internal information of
				  other SDN domains (provided that other domains allow to do
				  so), or obtaining domain-level information such as
				  inter-SDN-domain connectivity. None of these is applicable to
				  legacy ALTO client scenarios. As a result, ALTO server and
				  its protocol should be extended to support such scenarios.  
			  </t>
		  </section> <!-- downward-interaction -->
			  
		  <section anchor="upward-interaction" title="Upward Interaction"> 
			  <t> 
				  
				  The upward interaction is the information flow from ALTO
				  clients (i.e., SDN controllers) to ALTO servers. SDN
				  controllers open up the possibilities of conveniently
				  collecting network information and exporting such information
				  to ALTO servers. SDN controllers are at the best position to
				  collect network information.  
				  
				  More importantly, it is an inevitable requirement that SDN
				  controllers collect the information of the networking devices
				  in its domain.  
				  
				  <vspace blankLines="1"/> 
				  
				  In some scenarios, it is a requirement that information flow
				  from ALTO clients to ALTO servers. For instance, an SDN
				  domain could be dedicated to some of a carrier's certain
				  customers; the usage of such a domain gives privileged client
				  access. However, such a domain is an integral sub-network of
				  the carrier's network. In such a case, the ALTO server for
				  the carrier's network is not able to collect necessary
				  information in a scalable, manageable way. Even if we assume
				  that the ALTO server can automatically pull necessary
				  information directly from networking devices, the dedicated
				  domain may disallow the ALTO server to do so, because
				  customers who own and manage this domain may enforce
				  stringent privacy policies and disallow exporting information
				  externally.  The SDN controller is the best entity that can
				  facility the automation of information collection while at
				  the same time enforce the specific privacy policy.  
			  </t>
		  </section> <!-- upward-interaction -->
  
      </section>  <!-- SDNiALTO -->

      <section anchor="ALTOclisrv" title="Interactions between Legacy ALTO Clients and ALTO Servers">

         <t>With the existence of SDN, the way that legacy network applications
         (i.e., as legacy ALTO clients) interact with ALTO servers is also 
         different.</t>

         <t>In legacy ALTO client/server scenarios, ALTO clients obtain cost 
         maps from ALTO servers, with the implicit assumption that ALTO servers
         understand how the underlying network routes packets, which allows 
         ALTO servers to define or compute a cost metric associated with a 
         given route.</t>

         <t>However, with the introduction of SDN, such assumption may no longer
         hold, as SDN controllers may dynamically negotiate and determine a route
         between two end points (which may belong to two different SDN domains),
         especially when applications have specific requirements for network 
         resources (e.g.bandwidth, delay, etc). Thus, in order for applications
         to best utilize the network resources, the way that legacy ALTO clients
         communicate with ALTO servers should be adapted to SDN.</t>

      </section>  <!-- ALTOclisrv -->

   <section anchor="summ" title="Summary">

      <t>In the context of SDN, due to the specific and unique properties of 
      SDN domains, SDN controllers as ALTO clients are significantly different 
      from legacy ALTO clients, posing new requirements for the interactions 
      between ALTO clients and ALTO servers.</t> 

   </section>  <!-- summ -->
</section>  <!-- interact -->


<section anchor="info-flow" title="Information Flows"> 

	<t>We now further describe the two different information flows through two
		sets of use cases, one for the information flow from ALTO servers to
		ALTO clients, the other for the information flow from SDN controllers
		to ALTO servers.  </t>

   <section anchor="ctlFlow" title="Information Flow of SDN Controller">

      <t>A network may consist of multiple SDN domains. Note that due to operational
      or deployment concerns, there may exist networking devices that do not 
      belong to any SDN domain. In each SDN domain, the SDN controller is 
      responsible for the following tasks (only ALTO related tasks are 
      included below):
      <list style="symbols">
         <t>Collect fine-grain information from the networking devices it 
         manages. Such information could include, but not limited to, SDN 
         domain topology, link capacity, available bandwidth on each link, 
         links connected to external devices belonging to other SDN domains.</t>
         
         <t>Implement pre-defined domain-specific policies. Such policies could
         include, but not limited to, how resources should be allocated, how the
         collected information should combined and presented.</t>
         
         <t>Optionally aggregate the collected information for external view 
         purposes per its policies.</t>
         
         <t>Obtain cost maps at the granularity of SDN domains or obtain internal
         cost maps for specific domains (if available), consult for cross-domain
         data-forwarding plane recommendations from ALTO.</t>
         
         <t>Make (ALTO recommended) data/forwarding plane decisions based on
         the cost maps obtained from ALTO.</t>
      </list>
      </t>

   </section>  <!-- ctlFlow -->

   <section anchor="appFlow" title="Information Flow of Applications, SDN and ALTO">

      <t>We now describe three examples of complete information flows, which connect
      all key elements in an SDN.</t>

      <section anchor="SDNapp" title="SDN-aware Applications">

         <t> 
         <list style="symbols">
             <t>An application's end point sends a request for network resources
             to the SDN controller it belongs to (i.e., the SDN controller for 
             the SDN domain where this application's end point belongs to). The
             request should include the destination end point or the set of
             destination end points, and a set of requirements on network 
             resources (e.g., bandwidth)</t>
             
            <t>The SDN controller obtains an SDN-specific cost map from the 
            ALTO server (this step may occur independent of remaining steps)</t>
            
            <t>The SDN controller uses the cost map and negotiate one or many 
            path(s) with other SDN controllers (since the path may span across 
            multiple SDN domains, thus all SDN controllers of the involved domains
            should participate in setting up the paths)</t> 
            
            <t>The SDN controller responds to the requesting application's end 
            point.</t> 
            
            <t>If the requested path(s) are successfully set up, the application's
            end point starts to communicate with the destination end points.</t>
         </list>
         </t>
      </section>  <!-- SDNapp -->

      <section anchor="nSDNapp" title="SDN-unaware Applications">

	 <t>SDN-unaware applications do not directly communicate with SDN
	     controllers. Instead, they follow special packet formatting rules
	     to encode the SDN-specific requests, and the SDN capable
	     networking devices pick up these requests and forward them the SDN
	     controllers.</t> 

	 <t>The remaining information flow is similar to that of SDN-aware
	     applications, except that SDN controllers do not respond to the
	     requesting applications. Thus, when the requests cannot be
	     satisfied, SDN-unaware applications may suffer from packet losses,
	     due to networking devices process these applications' packets in a
	     best effort fashion.</t>

      </section>  <!-- nSDNapp -->

      <section anchor="legApp" title="Legacy Applications">

         <t>Legacy applications can be greatly simplified, as it is unnecessary 
         and is not helpful for them to directly communicate with ALTO servers 
         any more: 
         <list style="symbols">
            <t>An end point of a legacy application sends a packet to a known
            destination.</t>
            
            <t>A SDN capable networking device picks up the packet; however,
            if the path for the two end points has not been set up yet, the SDN
            controller will be consulted.</t>
            
            <t>The SDN controller obtains a cost map from the ALTO server (this
            step may occur independent of remaining steps).</t>
            
            <t>The SDN controller negotiate with other SDN controllers to set up 
            a best-effort path for the requesting end point.</t>
            
            <t>The forwarding rules for this path are pushed to all networking
            devices that are on the chosen path</t> <t>Communications between 
            the two end points continue; the forwarding rules may expire if the
            communication is tore down.</t> 
         </list> 
         </t> 
   
         <t>In this case, legacy applications are relieved from the complexity
         of dealing with the ALTO server using the ALTO protocol. ALTO-related
         intelligence, which fundamentally belongs to the network intelligence,
         is implemented in the network, rather than partly outside the network.</t>

      </section>  <!-- legApp -->
   </section>  <!-- appFlow -->

   <section anchor="info-flow-Summ" title="Summary">

      <t>It is worth noting that this architecture is fundamentally different
      from common ALTO use cases such as ALTO in CDN or data center (DC). The
      differences lie in that in the latter cases the components in question 
      (e.g., CDN or DC) are largely consumers of ALTO services, while in the 
      former case SDN domains are not only making decisions that may affect 
      ALTO and generating/aggregating information that ALTO needs, but also 
      the consumers of ALTO services. Furthermore, in the former case, SDN 
      domains are an integral part of the underlying network infrastructure 
      where their decisions could be treated as constraints for ALTO; however,
      in the latter cases, the components in question (e.g., CDN or DC) are 
      apparently not necessarily integral parts of the underlying network and
      their decisions could be treated as recommended outcomes suggested by
      ALTO.</t>

   </section>  <!-- info-flow-Summ -->
</section>  <!-- info-flow -->


<section anchor="upUse" title="Use Case for Upward Flow">

   <t>The upward flow delivers SDN domains' network information by SDN controllers 
   to the ALTO server. Each SDN controller is responsible for collecting, 
   aggregating, and submitting its own domain's network information to the ALTO
   server. Due to the possibility of some SDN domain being dedicated to certain 
   customers, we illustrate the upward flow in two use cases.</t>


   <section anchor="upUnres" title="Unrestrictive Information Exporting">

      <t>SDN domain controllers have to collect various network information from
      the networking devices they manage no matter if ALTO exists or not. The 
      reason is that an SDN controller may have to make decisions for allocating
      resources within its domain, and making such decisions need various network
      information. Since such information is readily collected and available, 
      an SDN controller could submit such information as is (or after simple
      processing) to the ALTO server.Take the available link bandwidth as an 
      example (available link bandwidth could be used as a measure of 
      "distance"). An SDN controller could periodically collect the available
      bandwidth on all links in its domain and submit it to the ALTO server. 
      However, such information should be annotated with the domain information
      (e.g., domain ID). By submitting such information, later other SDN 
      controllers may request for this domain's available link bandwidth 
      information.</t>

   </section>  <!-- upUnres -->

   <section anchor="upRest" title="Restrictive Information Exporting">

      <t>An SDN domain belonging to a carrier may be dedicated to certain
      customers of that carrier. In this case, the dedicated users of an SDN
      domain manage not only how resources should be allocated but also what
      information should be exported.</t>

      <t>A carrier may dedicate a set of small data centers (on multiple sites) 
      to its certain customer. These data centers are put under a single SDN 
      domain. The customer can manage the dedicated multi-site, small data 
      centers via the SDN controller. Periodically the SDN controller collects
      network information from all data centers.</t>

      <t>However, different than the former unrestrictive case, the customer may
      have stringent privacy policies and therefore decide to aggregate the 
      collected information before submitting to the ALTO server.</t> 

      <t>For instance, the customer may aggregate the information for a data 
      center network in the same site such that the data center network is
      shrunk into a single node; by doing so, the multi-site data center network
      is aggregated into a multi-node network topology, each node in the topology
      actually corresponds to a small data center in reality. The aggregated 
      network topology could be annotated with available link bandwidth
      information or other information that is collected and allowed to be 
      exported.</t>

      <t>The customer's information aggregation policy defines how the information
      should be pre-processed before exporting to the ALTO server. The main 
      purpose of aggregation is to protect privacy. As a result of information
      aggregation, the exported network information could be a logical topology
      (annotated with various network information, e.g., distance or cost) which
      is totally different from the physical topology.</t>

   </section>  <!-- upRest -->

   <section anchor="upAggreg" title="Information Aggregation">

      <t>Without SDN, ALTO defines cost maps for an aggregated view of the network 
      topology, where the granularity of aggregation is determined by the network 
      carrier and could be either coarse-grain or fine-grain.</t> 

      <t>However, with the introduction of SDN, such information aggregation could 
      be greatly simplified and should be policed based on the policies defined for
      each SDN domain. For instance, ALTO only needs to collect information from a 
      pre-defined set of SDN domain controllers, where the controllers determines 
      at what granularity they would like to aggregate the information and export 
      them. In addition, such aggregation is governed by the domain-specific policies,
      which defines not only the granularity of aggregation but also to whom such 
      aggregated information may be exposed.</t>

      <t>More specifically, an illustrative use case is as follows. SDN controllers
      collect fine-grain information and aggregate it periodically per their 
      policies. ALTO is configured to obtain the aggregated information from a
      set of SDN domain controllers and obtain possibly raw information from 
      networking devices (or the network operation center). ALTO then constructs
      a complete view of the overall network (an aggregated view of the network).
      SDN controllers obtain cost maps from ALTO and apply such maps when making
      data/forwarding plane decisions.</t>

      <t>Another illustrative use case is as follows. SDN controllers may choose 
      to export fine-grain information to ALTO. After it obtains the cost maps
      from ALTO, it could leverage the cost maps with greater details about their
      own domains and make informed decisions. However, SDN controllers should 
      not overload ALTO by exporting too much fine-grain information.</t>

   </section>  <!-- upAggreg -->
</section>  <!-- upUse -->


<section anchor="downUse" title="Use Case for Downward Flow">

   <t>We illustrate the use of downward flow through several use cases as 
   follows:</t>

   <t>Note that when the originating SDN domain's controller make decisions for 
   choosing path(s) and set up the path(s), each involved SDN domain controller 
   should map the overall decision to scoped decisions specifically for their 
   responsible domains.</t>
   

   <section anchor="SDNQoS" title="SDN-Aware QoS Metrics">

      <t>We use two use cases to describe SDN-aware QoS.
      When aggregating QoS information, SDN controllers or the information 
      aggregation policies should understand the semantics of each QoS metrics. 
      For instance, some metrics (e.g., delay) are additive, while some others are
      multiplicative (e.g., packet loss rate). The information aggregation policy 
      should be flexible enough to specify such details.</t>

      <section anchor="bod" title="On-Demand Bandwidth">

         <t>An SDN capable application / source end-point may request for 
         a certain amount of end-to-end bandwidth to a destination end-point 
         on the fly. The two end points in question should be in the same
         administrative domain, but they are not in the same SDN domain. The
         path(s) set up for such a request span across multiple SDN domains.</t>

         <t>The SDN controller of the source domain (i.e., the SDN domain where
         the source end-point is located), referred to as the source SDN 
         controller, should first obtain the cost maps from the ALTO server.
         Such cost maps are SDN-domain-specific, namely, the costs are defined 
         for pairs of SDN domains, rather than for pairs of end points as in 
         the legacy ALTO case.</t>

         <t>The source SDN domain controller should then determine path(s) for 
         the two end points based on the cost maps and associated information 
         obtained from ALTO. More specifically, the controller should: 
         <list style="symbols"> 
            <t>Compute a lowest-cost path at the SDN domain level using the 
            obtained SDN-domain-specific cost map.</t>
            
            <t>Contact the controllers of those SDN domains on the selected
            path, probing for the available bandwidth that could be dedicated
            to the requested session.</t>
            
            <t>Check if all of the selected path have sufficient combined 
            bandwidth that matches the required bandwidth.</t> 
            
            <t>if the combined bandwidth of all selected paths cannot match 
            the requirement, then go back to step 1 and select another 
            lowest-cost path (different than the already selected ones). 
            <list style="symbols">
               <t>if no path can be selected and the combined bandwidth does 
               not match the requirement, the request cannot be satisfied.</t>
            </list> 
            </t> 
            
            <t>if the combined bandwidth of all selected paths match the 
            requirement, then set up all selected paths by signaling all 
            involved SDN domain controllers. Note that the signaling protocol
            and how to set up paths are beyond the scope of this document.</t>
         </list>
         </t>

         <t>Data backup and migration among data centers, which typically 
         require bulk data transfers, is an example of on-demand bandwidth 
         use case. Data centers may be managed by one or multiple SDN domains;
         thus bulk data transfer could be thought of as bulk data transfer 
         among multiple SDN domains.</t>

      </section>  <!-- bod -->

      <section anchor="delay" title="Delay">

         <t>Similar to the preceding use case, applications may request for
         paths satisfying some certain QoS metrics, e.g., VoIP applications 
         may ask for paths with delay being lower than certain thresholds. This
         requires that ALTO cost maps embed such information, and that SDN 
         controller should export such information to ALTO.</t>

      </section>  <!-- delay -->
   </section>  <!-- SDNQoS -->

   <section anchor="CDN" title="Content Delivery Networks (CDN)">

      <t>Content Delivery Network (CDN) has been widely deployed to help 
      dissemination of content at the Internet scale. Network carries are 
      also deploying CDNs inside their own networks to improve the user 
      experiences of their customers. With the introduction of SDN, not only
      legacy CDN but also a new SDN-based CDN can be seamlessly implemented
      and integrated with the current network infrastructure.</t>

      <t>Here is an example of the flow of SDN-enabled CDN. Suppose that there 
      are a set of CDN servers deployed in a carrier's network and they are 
      willing to be managed by SDN. An equivalent class for each of the CDN 
      server is defined by either the CDN carrier or the network carrier (these
      two carriers can be the same). An equivalent class is a set of IP addresses,
      one for a CDN server, where if one can be used to fulfill requests for a
      specific content, then any server in this class can also be used to serve
      the same requests. In the extreme case, there is only one equivalent class
      for all CDN servers.</t>
   
      <t>Then the pre-defined equivalent classes are pushed to the SDN 
      controllers, which leverage such information to select CDN servers and
      set up paths for any end point to any such servers. 
      <list style="symbols">
         <t>A network client (e.g., an HTTP-based Web client) obtains the IP
         address, referred to as A, of one of the CDN servers in the carrier's
         network (e.g., by DNS queries).</t>
         
         <t>The client sends a first packet destined for A (for HTTP requests,
         this packet is a TCP SYN packet).</t>
         
         <t>An SDN capable networking device picks up the packet.</t>
         
         <t>If there are forwarding rules already set up for the communication
         between the requesting client and the destination A, then follow the 
         rules to forward the request packet.</t>
         
         <t>Otherwise, forward the request packet to the SDN controller of this
         domain.</t>
         
         <t>Once receiving a forwarded packet from a networking device, the
         SDN controller takes the following actions: 
         <list style="symbols"> 
            <t>Retrieves the equivalent class for the given destination A.</t>
            
            <t>Obtains a cost map from the ALTO server (this step could take
            place asynchronously).</t>
            
            <t>Ranks all CDN servers in the equivalent class according to the 
            cost map obtained from the ALTO server.</t>
            
            <t>Selects the best CDN server, referred to as B, based on the 
            above ranking.</t> 
            
            <t>Negotiates and sets up a best-effort path to the selected CDN 
            server with other controllers.</t>
            
            <t>Sets up forwarding rules for the path, and rewriting rules for 
            replacing the IP address of A with the IP address of B (so that the
            client is actually communicating with B, although it may think that
            it is communicating with A; however, which server it communicates is
            not important).</t> 
         </list> 
         </t> 
         
         <t>The request packet is forwarded to the chosen CDN server B, subject 
         to the forwarding rules and rewriting rules.</t> 
         
         <t>The client communicates with the CDN server B.</t>
         
	 <t>The path and associated forwarding/rewriting rules are expired when
	     the communication is torn down (this step is irrelevant to the
	     ALTO extension for SDN, therefore, it is out of scope).</t>

      </list> 
      </t>
      
      <t>However, the above use case has two limitations. First, it violates the TCP 
      semantics; namely, the client intends to and believes that it is 
      communicating with server A, but actually it is communicating with server B.
      Second, it has to rely on the capability of devices being able to rewrite 
      forwarding rules (e.g., use one IP address to replace another one in a 
      packet).  </t>
   
      <t>If the above two limitations become concerns, e.g., either TCP semantics
      should not be violated or rewriting is not available or both, the above
      SDN-enabled CDN use case can be implemented in similar way, with the help
      of a redirection server.</t> 
   
      <t>Below we describe the steps that are different:
	  <list style="symbols">
	      <t>A redirection server (or server farm), referred to as R, is set up for redirecting client requests.</t> 
	      
	      <t>Each SDN controller sets up path(s) to the given redirection server R.</t> 
	      
	      <t>Note that the redirection server could be an integral
		  component of an SDN controller (either collocated or
		  integrated), in which path(s) are not necessary.</t> 
	      
	      <t>Once receiving a forwarded packet from a networking device,
		  the SDN controller takes the following actions: 
		  
		  <list style="symbols"> 
		      <t>Retrieves the equivalent class for the given destination A.</t> 
		      
		      <t>Obtains a cost map from the ALTO server (this step
			  could take place asynchronously).</t> 
		      
		      <t>Ranks all CDN servers in the equivalent class
			  according to the cost map obtained from the ALTO
			  server.</t> 
		      
		      <t>Selects the best CDN server, referred to as B, based
			  on the above ranking.</t> 
		      
		      <t>Sends the information of the chosen CDN server, i.e.,
			  its IP address B, to the redirection server R.</t> 
		      
		      <t>Negotiates and sets up a best-effort path to the
			  redirection server R (if R is not integrated with the
			  SDN controller).</t>
		      
		      <t>Sets up forwarding rules for the path to R</t> 
		      
		      <t>Negotiates and sets up a best-effort path to the CDN
			  server B.</t> 
		      
		      <t>Sets up forwarding rules for the path to B.</t> 
	      </list> 
	  </t> 
          
          <t>The client communicates with the redirection server R.</t>
          
          <t>R sends an HTTP redirection packet to the client, redirecting 
          future requests to the CDN server B (which is notified by the SDN
          controller).</t>
          
          <t>The client communicates with the chosen CDN server B (note that the
          path to B has been already set up).</t>
       </list> 
       </t>

   </section>  <!-- CDN -->

   <section anchor="ICCDN" title="Information-Centric Content Delivery Networks (IC-CDN)">

      <t>Information-Centric Networking (ICN) is a "host-to-information" 
      communication model, different from the legacy "host-to-host" model 
      implemented by the Internet. Content Delivery Network (CDN) is more 
      of a "host-to-information" model (i.e., CDNs can be treated as a special
      instance of ICN), but implemented in the "host-to-host" model, due to 
      the fact that the current semantics provided by the Internet only support
      the "host-to-host" model.</t>

      <t> With the introduction of SDN, CDNs can be converted into an 
      information-centric networking implementation: 
      <list style="symbols"> 
         <t>A CDN client sends a request for a specific content.</t>
         
         <t>The request packet is formatted per the CDN in SDN specification
         (beyond the scope of this draft), containing: 
         <list style="symbols"> 
            <t>the requested content name in the packet.</t>
            
            <t>destination (a specific anycast IP address) which is 
            reserved for legacy applications to invoke SDN capabilities.</t>
            
            <t>(optional) QoS requirements (e.g., prefer fast/local servers 
            vs. slow/remote servers, demands are elastic or not).</t>
         </list>
         </t>
         
         <t>An SDN capable networking device picks up the request packet.</t>
         
         <t>If there are forwarding rules set up for this content request
         already, then follow the rules to forward the request packet, and 
         terminate this.</t> 
         
         <t>Otherwise, forward the request packet to the SDN controller for 
         this domain.</t>
         
         <t>The SDN controller communicates with the CDN's name directory to 
         look up possible CDN servers that can satisfy the request.</t>
         
         <t>The SDN controller obtains a cost map from the ALTO server.</t>
         
         <t>The SDN controller applies the cost map to select the best CDN 
         server per the QoS requirements if specified in the request.</t>
         
         <t>The SDN controller negotiate the path to the selected CDN server
         with other controllers.</t>
         
         <t>The SDN controllers that along the chosen path set up the path,
         and push the forwarding rule(s) for this chosen path to all networking
         devices that are involved.</t>
         
         <t>The request packet is forwarded to the chosen CDN server.</t>
         
         <t>Data packets flow back to the CDN client.</t> 
      </list> 
      </t>

      <t>In this use case, the CDN clients could be modified to send the 
      "information-centric" request. However, in a realistic implementation,
      neither the CDN clients nor the CDN servers have to be significantly
      modified (e.g., CDN redirection could be leveraged to implement the 
      above work flow).</t>

   </section>  <!-- ICCDN -->
</section>  <!-- downUse -->
   

<section anchor="finConc" title="Conclusions"> 
	
	<t>
		
		In this draft, we identify the fundamental differences between legacy
		ALTO client/server and ALTO client/server with the existence of SDN.
		The introduction of SDN fundamentally changes the way that the ALTO
		works.  We present the Vertical Architecture and the Horizontal
		Architecture to allow coherent coexistence of SDN and ALTO. We believe
		that the Vertical Architecture allows better division, management,
		flexibility, privacy control and long-term evolution of the network.  
		
		Therefore we mainly focus on the Vertical Architecture in this draft.
		We also define the main interactions and information flows, and present
		a set of use cases to illustrate how we extend ALTO to support SDN, in
		the Vertical Architecture.

	</t>

</section>  <!-- finConc -->


<section anchor="contrib" title="Contributors">

   <t>The authors would like to thank Vijay K. Gurbani for his many detailed reviews 
   and helpful assistance on this draft.</t>
     
   <t>Vijay K. Gurbani
   <vspace blankLines="0"/>
   Bell Laboratories, Alcatel-Lucent
   <vspace blankLines="0"/>
   1960 Lucent Lane, Rm. 9C-533
   <vspace blankLines="0"/>
   Naperville, IL 60566
   <vspace blankLines="0"/>
   USA
   <vspace blankLines="1"/>
   Email: vkg AT (acm.org,bell-labs.com)
   </t>
   
</section>


<section anchor="ack" title="Acknowledgements">

   <t>The authors would like to thank Christos Kolias for his reviews and
       feedbacks, and thank Aditi Vira for editing the draft.</t>

</section>


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

   <t>TBD.</t>

</section>  <!-- secur -->


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

   <t>This document makes no specific request of IANA.</t>
   
   <t>Note to RFC Editor: this section may be removed on publication as an RFC.</t>

</section>  <!-- IANA -->

  </middle>

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

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

<references title="Informative References">

   <reference anchor="abstr">
      <front>
          <title>Abstracting network state in Software Defined Networks (SDN) for rendezvous services, IEEE International Conference on Communications (ICC) Workshop on Software Defined Networks (SDN)</title>
         <author initials="V.K." surname="Gurbani" fullname="V.K. Gurbani">
            <organization>Bell Laboratories, Alcatel-Lucent</organization>
         </author>
         <author initials="M." surname="Scharf" fullname="M. Scharf">
            <organization></organization>
         </author>
         <author initials="T.V." surname="Lakshman" fullname="T.V. Lakshman">
            <organization></organization>
         </author>
         <author initials="V." surname="Hilt" fullname="V. Hilt">
            <organization></organization>
         </author>
         <date month="June" year="2012"/>
      </front>
   </reference>

   <reference anchor="onix">
       <front>
          <title>Onix: A Distributed Control Platform for Large-scale Production Networks", Proceedings of the 9th USENIX Symposium on Operating Systems Design and Implementation (OSDI 10), Vancouver, Canada, pp. 351-364</title>
         <author initials="T." surname="Koponen" fullname="T. Koponen">
      <organization></organization>
         </author>
         <author initials="M." surname="Casado" fullname="M. Casado">
            <organization></organization>
         </author>
         <author initials="N." surname="Gude" fullname="N. Gude">
            <organization></organization>
         </author>
         <author initials="J." surname="Stribling" fullname="J. Stribling">
            <organization></organization>
         </author>
         <author initials="L." surname="Poutievski" fullname="L. Poutievski">
            <organization></organization>
         </author>
         <author initials="M." surname="Zhu" fullname="M. Zhu">
            <organization></organization>
         </author>
         <author initials="R." surname="Ramanathan" fullname="R. Ramanathan">
            <organization></organization>
         </author>
         <author initials="Y." surname="Iwata" fullname="Y. Iwata">
            <organization></organization>
         </author>
         <author initials="H." surname="Inoue" fullname="H. Inoue">
            <organization></organization>
         </author>
         <author initials="T." surname="Hama" fullname="T. Hama">
            <organization></organization>
         </author>
         <author initials="S." surname="Shenker" fullname="S. Shenker">
            <organization></organization>
         </author>
         <date month="October" year="2010"/>
      </front>
   </reference>

</references>

  </back>
</rfc>

PAFTECH AB 2003-20262026-04-24 09:57:12