Internet Engineering Task Force                          Marc Lasserre
    Internet Draft                                            Florin Balus
    Intended status: Informational                          Alcatel-Lucent
    Expires: March 2013
                                                              Thomas Morin
                                                     France Telecom Orange

                                                               Nabil Bitar

                                                             Yakov Rekhter

                                                        September 11,

                                                          October 19, 2012

                      Framework for DC Network Virtualization

    Status of this Memo

       This Internet-Draft is submitted in full conformance with the
       provisions of BCP 78 and BCP 79.

       Internet-Drafts are working documents of the Internet Engineering
       Task Force (IETF).  Note that other groups may also distribute
       working documents as Internet-Drafts. The list of current Internet-
       Drafts is at

       Internet-Drafts are draft documents valid for a maximum of six
       months and may be updated, replaced, or obsoleted by other documents
       at any time.  It is inappropriate to use Internet-Drafts as
       reference material or to cite them other than as "work in progress."

       This Internet-Draft will expire on March 11, April 19, 2013.

    Copyright Notice

       Copyright (c) 2012 IETF Trust and the persons identified as the
       document authors. All rights reserved.

       This document is subject to BCP 78 and the IETF Trust's Legal
       Provisions Relating to IETF Documents
       ( in effect on the date of
       publication of this document. Please review these documents
       carefully, as they describe your rights and restrictions with
       respect to this document. Code Components extracted from this
       document must include Simplified BSD License text as described in
       Section 4.e of the Trust Legal Provisions and are provided without
       warranty as described in the Simplified BSD License.


       Several IETF drafts relate to the use of overlay networks to support
       large scale virtual data centers. This draft provides a framework
       for Network Virtualization over L3 (NVO3) and is intended to help
       plan a set of work items in order to provide a complete solution
       set. It defines a logical view of the main components with the
       intention of streamlining the terminology and focusing the solution

    Table of Contents

       1. Introduction.................................................3 Introduction................................................3
          1.1. Conventions used in this document.......................4
          1.2. General terminology.....................................4
          1.3. DC network architecture.................................6
          1.4. Tenant networking view..................................7
       2. Reference Models.............................................8 Models............................................8
          2.1. Generic Reference Model.................................8
          2.2. NVE Reference Model....................................10
          2.3. NVE Service Types......................................11 Types......................................12
             2.3.1. L2 NVE providing Ethernet LAN-like service........11 service.........12
             2.3.2. L3 NVE providing IP/VRF-like service..............11 service..............12
       3. Functional components.......................................11 components.......................................12
          3.1. Generic service virtualization components..............12
             3.1.1. Virtual Access Points (VAPs)......................12 (VAPs)......................13
             3.1.2. Virtual Network Instance (VNI)....................12 (VNI)....................13
             3.1.3. Overlay Modules and VN Context....................13
             3.1.4. Tunnel Overlays and Encapsulation options.........14 options..........14
             3.1.5. Control Plane Components..........................14
    Auto-provisioning/Service discovery.............14 Distributed vs Centralized Control Plane.........15
    Auto-provisioning/Service discovery.............15
    Address advertisement and tunnel mapping........15
    Tunnel management...............................15 management...............................17
          3.2. Multi-homing..........................................17
          3.3. Service Overlay Topologies.............................16 Topologies.............................18
       4. Key aspects of overlay networks.............................16 networks.............................18
          4.1. Pros & Cons............................................16 Cons...........................................18
          4.2. Overlay issues to consider.............................17 consider.............................19
             4.2.1. Data plane vs Control plane driven................17 driven................19
             4.2.2. Coordination between data plane and control plane..18 plane..20
             4.2.3. Handling Broadcast, Unknown Unicast and Multicast (BUM)
             4.2.4. Path MTU..........................................19 MTU.........................................21
             4.2.5. NVE location trade-offs...........................19 trade-offs...........................21
             4.2.6. Interaction between network overlays and underlays.20 underlays.22
       5. Security Considerations.....................................21 Considerations.....................................23
       6. IANA Considerations.........................................21 Considerations........................................23
       7. References..................................................21 References.................................................23
          7.1. Normative References...................................21 References...................................23
          7.2. Informative References.................................21 References.................................23
       8. Acknowledgments.............................................22 Acknowledgments............................................24

    1. Introduction

       This document provides a framework for Data Center Network
       Virtualization over L3 tunnels. This framework is intended to aid in
       standardizing protocols and mechanisms to support large scale
       network virtualization for data centers.

       Several IETF drafts relate to the use of overlay networks for data

       [NVOPS] defines the rationale for using overlay networks in order to
       build large data center networks. The use of virtualization leads to
       a very large number of communication domains and end systems to cope

       [OVCPREQ] describes the requirements for a control plane protocol
       required by overlay border nodes to exchange overlay mappings.

       This document provides reference models and functional components of
       data center overlay networks as well as a discussion of technical
       issues that have to be addressed in the design of standards and
       mechanisms for large scale data centers.

    1.1. Conventions used in this document

       The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
       "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
       document are to be interpreted as described in RFC-2119 [RFC2119].

       In this document, these words will appear with that interpretation
       only when in ALL CAPS. Lower case uses of these words are not to be
       interpreted as carrying RFC-2119 significance.

    1.2. General terminology

       This document uses the following terminology:

       NVE: Network Virtualization Edge. It is a network entity that sits
       on the edge of the NVO3 network. It implements network
       virtualization functions that allow for L2 and/or L3 tenant
       separation and for hiding tenant addressing information (MAC and IP
       addresses). An NVE could be implemented as part of a virtual switch
       within a hypervisor, a physical switch or router, a Network Service
       Appliance or even be embedded within an End Station.

       VN: Virtual Network. This is a virtual L2 or L3 domain that belongs
       to a tenant.

       VNI: Virtual Network Instance. This is one instance of a virtual
       overlay network. Two Virtual Networks are isolated from one another
       and may use overlapping addresses.

       Virtual Network Context or VN Context: Field that is part of the
       overlay encapsulation header which allows the encapsulated frame to
       be delivered to the appropriate virtual network endpoint by the
       egress NVE. The egress NVE uses this field to determine the
       appropriate virtual network context in which to process the packet.
       This field MAY be an explicit, unique (to the administrative domain)
       virtual network identifier (VNID) or MAY express the necessary
       context information in other ways (e.g. a locally significant

       VNID:  Virtual Network Identifier. In the case where the VN context
       has global significance, this is the ID value that is carried in
       each data packet in the overlay encapsulation that identifies the
       Virtual Network the packet belongs to.

       Underlay or Underlying Network: This is the network that provides
       the connectivity between NVEs. The Underlying Network can be
       completely unaware of the overlay packets. Addresses within the
       Underlying Network are also referred to as "outer addresses" because
       they exist in the outer encapsulation. The Underlying Network can
       use a completely different protocol (and address family) from that
       of the overlay.

       Data Center (DC): A physical complex housing physical servers,
       network switches and routers, Network Service Appliances and
       networked storage. The purpose of a Data Center is to provide
       application and/or compute and/or storage services. One such service
       is virtualized data center services, also known as Infrastructure as
       a Service.

       Virtual Data Center or Virtual DC: A container for virtualized
       compute, storage and network services. Managed by a single tenant, a
       Virtual DC can contain multiple VNs and multiple Tenant End Systems that
       are connected to one or more of these VNs.

       VM: Virtual Machine. Several Virtual Machines can share the
       resources of a single physical computer server using the services of
       a Hypervisor (see below definition).

       Hypervisor: Server virtualization software running on a physical
       compute server that hosts Virtual Machines. The hypervisor provides
       shared compute/memory/storage and network connectivity to the VMs
       that it hosts. Hypervisors often embed a Virtual Switch (see below).

       Virtual Switch: A function within a Hypervisor (typically
       implemented in software) that provides similar services to a
       physical Ethernet switch.  It switches Ethernet frames between VMs'
       virtual NICs within the same physical server, or between a VM and a
       physical NIC card connecting the server to a physical Ethernet
       switch. It also enforces network isolation between VMs that should
       not communicate with each other.

       Tenant: A customer who consumes virtualized data center services
       offered by In a DC, a cloud service provider. A single tenant may consume one refers to a customer that could an
       organization within an enterprise, or more Virtual Data Centers hosted by the same cloud service
       provider. an enterprise with a set of DC
       compute, storage and network resources associated with it.

       Tenant End System: It defines an end A physical or virtual system that can play the role
       of a particular tenant,
       which can be for instance host, or a virtual machine (VM), forwarding element such as a non-virtualized
       server, or router, switch,
       firewall, etc. It belongs to a single tenant and connects to one or
       more VNs of that tenant.

       End device: A physical appliance. system to which networking service is
       provided. Examples include hosts (e.g. server or server blade),
       storage systems (e.g. file servers, iSCSI storage systems) and
       network devices (e.g. firewall, load-balancer, IPSec gateway). An
       end device may include internal networking functionality that
       interconnects the device's components (e.g. virtual switches that
       interconnects VMs running on the same server). NVE functionality may
       be implemented as part of that internal networking.

       ELAN: MEF ELAN, multipoint to multipoint Ethernet service

       EVPN: Ethernet VPN as defined in [EVPN]

    1.3. DC network architecture

       A generic architecture for Data Centers is depicted in Figure 1:

                                  ,'           `.
                                 (  IP/MPLS WAN )
                                  `.           ,'
                                 +--+--+   +-+---+
                                 |DC GW|+-+|DC GW|
                                 +-+---+   +-----+
                                    |       /
                                    .--. .--.
                                  (    '    '.--.
                                .-.' Intra-DC     '
                               (     network      )
                                (             .'-'
                                 '--'._.'.    )\ \
                                 / /     '--'  \ \
                                / /      | |    \ \
                          +---+--+   +-`.+--+  +--+----+
                          | ToR  |   | ToR  |  |  ToR  |
                          +-+--`.+   +-+-`.-+  +-+--+--+
                           /     \   .'    /    \   /       \   .'     `.
                        __/_      _i./       i./_      \  /      \ /_       _\__
                 '--------'   '--------'   '--------'   '--------'
                 :  End   :   :  End   :   :  End   :   :  End   :
                 : Device :   : Device :   : Device :   : Device :
                 '--------'   '--------'   '--------'   '--------'

                 Figure 1 : A Generic Architecture for Data Centers

       An example of multi-tier DC network architecture is presented in
       this figure. It provides a view of physical components inside a DC.

       A cloud network is composed of intra-Data Center (DC) networks and
       network services, and, inter-DC network and network connectivity
       services. Depending upon the scale, DC distribution, operations
       model, Capex and Opex aspects, DC networking elements can act as
       strict L2 switches and/or provide IP routing capabilities, including
       also service virtualization.

       In some DC architectures, it is possible that some tier layers
       provide L2 and/or L3 services, are collapsed, and that Internet
       connectivity, inter-DC connectivity and VPN support are handled by a
       smaller number of nodes. Nevertheless, one can assume that the
       functional blocks fit with the architecture above.

       The following components can be present in a DC:

          o End Device: a DC resource to which the networking service is
            provided. End Device may be a compute resource (server or
            server blade), storage component or a network appliance
            (firewall, load-balancer, IPsec gateway). Alternatively, the
            End Device may include software based networking functions used
            to interconnect multiple hosts. An example of soft networking
            is the virtual switch in the server blades, used to
            interconnect multiple virtual machines (VMs). End Device may be
            single or multi-homed to the Top of Rack switches (ToRs).

          o Top of Rack (ToR): Hardware-based Ethernet switch aggregating
            all Ethernet links from the End Devices in a rack representing
            the entry point in the physical DC network for the hosts. ToRs
            may also provide routing functionality, virtual IP network
            connectivity, or Layer2 tunneling over IP for instance. ToRs
            are usually multi-homed to switches in the Intra-DC network.
            Other deployment scenarios may use an intermediate Blade Switch
            before the ToR or an EoR (End of Row) switch to provide similar
            function as a ToR.

          o Intra-DC Network: High capacity network composed of core
            switches aggregating multiple ToRs. Core switches are usually
            Ethernet switches but can also support routing capabilities.

          o DC GW: Gateway to the outside world providing DC Interconnect
            and connectivity to Internet and VPN customers. In the current
            DC network model, this may be simply a Router connected to the
            Internet and/or an IPVPN/L2VPN PE. Some network implementations
            may dedicate DC GWs for different connectivity types (e.g., a
            DC GW for Internet, and another for VPN).

       Note that End Devices may be single or multi-homed to ToRs.

    1.4. Tenant networking view

       The DC network architecture is used to provide L2 and/or L3 service
       connectivity to each tenant. An example is depicted in Figure 2:

                         +----- L3 Infrastructure ----+
                         |                            |
                          ,--+-'.                      ;--+--.
                      ,--+--.                      ,--+--.
                .....( Rtr1  )......              .              ( Rtr2  )
                |    '-----'     `-----'      |               '-----'               `-----'
                |     Tenant1      |LAN12      Tenant1|
                |LAN11         ....|........          |LAN13
          ..............        |        |     '':'''''''':'
                  ,'.      ,'.      ,+.      ,+.     ,'.      ,'.
                 (VM     ..............
             |        |         |        |       |        |
            ,-.      ,-.       ,-.      ,-.     ,-.      ,-.
           (VM )....(VM )     (VM )... (VM )   (VM )....(VM )
            `-'      `-'       `-'      `-'     `-'      `-'

            Figure 2 : Logical Service connectivity for a single tenant

       In this example one or more L3 contexts and one or more LANs (e.g.,
       one per application type) running on DC switches are assigned for DC
       tenant 1.

       For a multi-tenant DC, a virtualized version of this type of service
       connectivity needs to be provided for each tenant by the Network
       Virtualization solution.

    2. Reference Models

    2.1. Generic Reference Model

       The following diagram shows a DC reference model for network
       virtualization using Layer3 overlays where edge devices NVEs provide a logical
       interconnect between Tenant End Systems that belong to specific tenant

             +--------+                                    +--------+
             | Tenant |                                  | Tenant |
             |  End +--+                           +---|  End                            +----| Tenant |
             | System |  |                           |                           (')   | System |
             +--------+  |    ...................    |   (   )  +--------+
                         |  +-+--+           +--+-+  |  (_)
                         |  | NV |           | NV |   |
                         +--|Edge|           |Edge|--+           |Edge|---+
                            +-+--+           +--+-+
                            / .                 .
                           /  .   L3 Overlay   .  \ +--+-++--------+
             +--------+   /   .    Network     .   \     +--------+   | NV || Tenant +--+    .                 .    +----| Tenant | |  End
             | Tenant +--+    .                 .         |  End              |Edge|| System |
             | System |       .    +----+       .         | System |    +--+-++--------+
             +--------+       .....| NV |........         +--------+
                             |               |
                         +--------+      +--------+
                         | Tenant |      |  End Tenant |
                         | System |      | System |
                         +--------+      +--------+

          Figure 3 : Generic reference model for DC network virtualization
                           over a Layer3 infrastructure

       A Tenant System can be attached to a Network Virtualization Edge
       (NVE) node in several ways:

         - locally, by being co-located i.e. resident in the same device

         - remotely, via a point-to-point connection or a switched network
         (e.g. Ethernet)

       When an NVE is local, the state of Tenant Systems can be provided
       without protocol assistance. For instance, the operational status of
       a VM can be communicated via a local API. When an NVE is remote, the
       state of Tenant Systems needs to be exchanged via a data or control
       plane protocol, or via a management entity.

       The functional components in this picture do not necessarily map
       directly with the physical components described in Figure 1.

       For example, an End Device can be a server blade with VMs and
       virtual switch, i.e. the VM is the Tenant End System and the NVE
       functions may be performed by the virtual switch and/or the
       hypervisor. In this case, the Tenant System and NVE function are co-

       Another example is the case where an End Device can be a traditional
       physical server (no VMs, no virtual switch), i.e. the server is the
       Tenant End System and the NVE functions function may be performed by the ToR.
       Other End Devices in this category are Physical Network Appliances
       or Storage Systems.

       A Tenant End System attaches to a Network Virtualization Edge (NVE)
       node, either directly or via a switched network (typically

       The NVE implements network virtualization functions that allow for
       L2 and/or L3 tenant separation and for hiding tenant addressing
       information (MAC and IP addresses), tenant-related control plane
       activity and service contexts from the Routed Backbone nodes.

       Core nodes utilize L3 techniques to interconnect NVE nodes in
       support of the overlay network. These devices perform forwarding
       based on outer L3 tunnel header, and generally do not maintain per
       tenant-service state albeit some applications (e.g., multicast) may
       require control plane or forwarding plane information that pertain
       to a tenant, group of tenants, tenant service or a set of services
       that belong to one or more tunnels. When such tenant or tenant-
       service related information is maintained in the core, overlay
       virtualization provides knobs to control that information.

    2.2. NVE Reference Model

       The NVE is composed of a tenant service Virtual Network instance that Tenant End
       Systems interface with and an overlay module that provides tunneling
       overlay functions (e.g. encapsulation/decapsulation of tenant
       traffic from/to the tenant forwarding instance, tenant
       identification and mapping, etc), as described in figure 4:

                          +------- L3 Network ------+
                          |                         |
                          |       Tunnel Overlay    |
             +------------+---------+       +---------+------------+
             | +----------+-------+ |       | +---------+--------+ |
             | |  Overlay Module  | |       | |  Overlay Module  | |
             | +---------+--------+ |       | +---------+--------+ |
             |           |VN context|       | VN context|          |
             |           |          |       |           |          |
             |  +--------+-------+  |       |  +--------+-------+  |
             |  | |VNI|   .  |VNI|  |       |  | |VNI|   .  |VNI|  |
        NVE1 |  +-+------------+-+  |       |  +-+-----------+--+  | NVE2
             |    |   VAPs     |    |       |    |    VAPs   |     |
             +----+------------+----+       +----+------------+----+       +----+-----------+-----+
                  |            |                 |           |
                  |            |     Tenant      |           |
                  |            |   Service IF    |           |
                 Tenant End Systems                 Tenant End Systems

                  Figure 4 : Generic reference model for NV Edge

       Note that some NVE functions (e.g. data plane and control plane
       functions) may reside in one device or may be implemented separately
       in different devices.

       For example, the NVE functionality could reside solely on the End
       Devices, on the ToRs or on both the End Devices and the ToRs. In the
       latter case we say that the the End Device NVE component acts as the NVE
       Spoke, and ToRs act as NVE hubs. Tenant End Systems will interface with the tenant service instances
       VNIs maintained on the NVE spokes, and tenant service instances VNIs maintained on the NVE
       spokes will interface with the tenant service instances VNIs maintained on the NVE hubs.

    2.3. NVE Service Types

       NVE components may be used to provide different types of virtualized
       service connectivity. This section defines the service types and
       associated attributes

    2.3.1. L2 NVE providing Ethernet LAN-like service

       L2 NVE implements Ethernet LAN emulation (ELAN), an Ethernet based
       multipoint service where the Tenant End Systems appear to be
       interconnected by a LAN environment over a set of L3 tunnels. It
       provides per tenant virtual switching instance with MAC addressing
       isolation and L3 tunnel encapsulation across the core.

    2.3.2. L3 NVE providing IP/VRF-like service

       Virtualized IP routing and forwarding is similar from a service
       definition perspective with IETF IP VPN (e.g., BGP/MPLS IPVPN and
       IPsec VPNs). It provides per tenant routing instance with addressing
       isolation and L3 tunnel encapsulation across the core.

    3. Functional components

       This section breaks down the Network Virtualization architecture
       into functional components to make it easier to discuss solution
       options for different modules.

       This version of the document gives an overview of generic functional
       components that are shared between L2 and L3 service types. Details
       specific for each service type will be added in future revisions.

    3.1. Generic service virtualization components

       A Network Virtualization solution is built around a number of
       functional components as depicted in Figure 5:

                         +------- L3 Network ------+
                         |                         |
                         |       Tunnel Overlay    |
            +------------+--------+       +--------+------------+
            | +----------+------+ |       | +------+----------+ |
            | | Overlay Module  | |       | | Overlay Module  | |
            | +--------+--------+ |       | +--------+--------+ |
            |          |VN Context|       |          |VN Context|
            |          |          |       |          |          |
            |  +-------+-------+  |       |  +-------+-------+  |
            |  ||VNI| ... |VNI||  |       |  ||VNI| ... |VNI||  |
       NVE1 |  +-+-----------+-+  |       |  +-+-----------+-+  | NVE2
            |    |   VAPs    |    |       |    |   VAPs    |    |
            +----+-----------+----+       +----+-----------+----+
                 |           |                 |           |
                 |           |     Tenant      |           |
                 |           |   Service IF    |           |
                Tenant End Systems                Tenant End Systems

                  Figure 5 : Generic reference model for NV Edge

    3.1.1. Virtual Access Points (VAPs)

       Tenant End Systems are connected to the VNI Instance through Virtual
       Access Points (VAPs).

       The VAPs can be in reality physical ports on a
       ToR or virtual ports identified through
       logical interface identifiers (VLANs, internal VSwitch Interface ID
       leading to a VM).

    3.1.2. Virtual Network Instance (VNI)

       The VNI represents a set of configuration attributes defining access
       and tunnel policies and (L2 and/or L3) forwarding functions.

       Per tenant FIB tables and control plane protocol instances are used
       to maintain separate private contexts between tenants. Hence tenants
       are free to use their own addressing schemes without concerns about
       address overlapping with other tenants.

    3.1.3. Overlay Modules and VN Context

       Mechanisms for identifying each tenant service are required to allow
       the simultaneous overlay of multiple tenant services over the same
       underlay L3 network topology. In the data plane, each NVE, upon
       sending a tenant packet, must be able to encode the VN Context for
       the destination NVE in addition to the L3 tunnel source address
       identifying the source NVE and the tunnel destination L3 address
       identifying the destination NVE. This allows the destination NVE to
       identify the tenant service instance and therefore appropriately
       process and forward the tenant packet.

       The Overlay module provides tunneling overlay functions: tunnel
       initiation/termination, encapsulation/decapsulation of frames from
       VAPs/L3 Backbone and may provide for transit forwarding of IP
       traffic (e.g., transparent tunnel forwarding).

       In a multi-tenant context, the tunnel aggregates frames from/to
       different VNIs. Tenant identification and traffic demultiplexing are
       based on the VN Context (e.g. VNID).

       The following approaches can been considered:

          o One VN Context per Tenant: A globally unique (on a per-DC
            administrative domain) VNID is used to identify the related
            Tenant instances. An example of this approach is the use of
            IEEE VLAN or ISID tags to provide virtual L2 domains.

          o One VN Context per VNI: A per-tenant local value is
            automatically generated by the egress NVE and usually
            distributed by a control plane protocol to all the related
            NVEs. An example of this approach is the use of per VRF MPLS
            labels in IP VPN [RFC4364].

          o One VN Context per VAP: A per-VAP local value is assigned and
            usually distributed by a control plane protocol. An example of
            this approach is the use of per CE-PE MPLS labels in IP VPN

       Note that when using one VN Context per VNI or per VAP, an
       additional global identifier may be used by the control plane to
       identify the Tenant context.

    3.1.4. Tunnel Overlays and Encapsulation options

       Once the VN context is added to the frame, a L3 Tunnel encapsulation
       is used to transport the frame to the destination NVE. The backbone
       devices do not usually keep any per service state, simply forwarding
       the frames based on the outer tunnel header.

       Different IP tunneling options (GRE/L2TP/IPSec) and tunneling
       options (BGP VPN, PW, VPLS) are available for both Ethernet and IP

    3.1.5. Control Plane Components

       Control plane components may be used to provide the following

          . Auto-provisioning/Service discovery

          . Address advertisement and tunnel mapping

          . Tunnel management

       A control plane component can be an on-net control protocol or a
       management control entity. Auto-provisioning/Service discovery

       NVEs must Distributed vs Centralized Control Plane

       A control/management plane entity can be able to select centralized or distributed.
       Both approaches have been used extensively in the appropriate VNI for each Tenant End
       System. This is based on state information that past. The routing
       model of the Internet is often provided by
       external entities. For example, in a VM environment, this
       information is provided by compute management systems, since good example of a distributed approach.
       Transport networks have usually used a centralized approach to
       manage transport paths.

       It is also possible to combine the two approaches i.e. using a
       hybrid model. A global view of network state can have many benefits
       but it does not preclude the use of distributed protocols within the
       network. Centralized controllers provide a facility to maintain
       global and distribute that state to the network which in combination
       with distributed protocols can aid in achieving greater network
       efficiencies, improve reliability and robustness. Domain and/or
       deployment specific constraints define the balance between
       centralized and distributed approaches.

       On one hand, a control plane module can reside in every NVE. This is
       how routing control plane modules are implemented in routers. At the
       same time, an external controller can manage a group of NVEs via an
       agent sitting in each NVE. This is how an SDN controller could
       communicate with the nodes it controls, via OpenFlow for instance.

       In the case where a centralized control plane is preferred, the
       controller will need to be distributed to more than one node for
       redundancy. Depending upon the size of the DC domain, hence the
       number of NVEs to manage, it should be possible to use several
       external controllers. Inter-controller communication will thus be
       necessary for scalability and redundancy. Auto-provisioning/Service discovery

       NVEs must be able to select the appropriate VNI for each Tenant
       System. This is based on state information that is often provided by
       external entities. For example, in a VM environment, this
       information is provided by compute management systems, since these
       are the only entities that have visibility on which VM belongs to
       which tenant.

       A mechanism for communicating this information between Tenant End
       Systems and the local NVE is required. As a result the VAPs are
       created and mapped to the appropriate Tenant Instance. VNI.

       Depending upon the implementation, this control interface can be
       implemented using an auto-discovery protocol between Tenant End Systems
       and their local NVE or through management entities.

       When a protocol is used, appropriate security and authentication
       mechanisms to verify that Tenant End System information is not spoofed
       or altered are required. This is one critical aspect for providing
       integrity and tenant isolation in the system.

       Another control plane protocol can also be used to advertize NVE
       tenant service instance (tenant and service type provided to the
       supported VNs to other NVEs. Alternatively, management control
       entities can also be used to perform these functions. Address advertisement and tunnel mapping

       As traffic reaches an ingress NVE, a lookup is performed to
       determine which tunnel the packet needs to be sent to. It is then
       encapsulated with a tunnel header containing the destination address
       of the egress overlay node. Intermediate nodes (between the ingress
       and egress NVEs) switch or route traffic based upon the outer
       destination address.

       One key step in this process consists of mapping a final destination
       address to the proper tunnel. NVEs are responsible for maintaining
       such mappings in their lookup tables. Several ways of populating
       these lookup tables are possible: control plane driven, management
       plane driven, or data plane driven.

       When a control plane protocol is used to distribute address
       advertisement and tunneling information, the auto-
       provisioning/Service discovery could be accomplished by the same
       protocol. In this scenario, the auto-provisioning/Service discovery
       could be combined with (be inferred from) the address advertisement
       and tunnel mapping. Furthermore, a control plane protocol that
       carries both MAC and IP addresses eliminates addresses eliminates the need for ARP, and
       hence addresses one of the issues with explosive ARP handling. Tunnel management

       A control plane protocol may be required to exchange tunnel state
       information. This may include setting up tunnels and/or providing
       tunnel state information.

       This applies to both unicast and multicast tunnels.

       For instance, it may be necessary to provide active/standby status
       information between NVEs, up/down status information,
       pruning/grafting information for multicast tunnels, etc.

    3.2. Multi-homing

       Multi-homing techniques can be used to increase the reliability of
       an nvo3 network. It is also important to ensure that physical
       diversity in an nvo3 network is taken into account to avoid single
       points of failure.

       Multi-homing can be enabled in various nodes, from tenant systems
       into TORs, TORs into core switches/routers, and core nodes into DC

       The nvo3 underlay nodes (i.e. from NVEs to DC GWs) rely on IP
       routing as the means to re-route traffic upon failures and/or ECMP

       Tenant systems can either be L2 or L3 nodes. In the former case
       (L2), techniques such as LAG or STP for instance can be used. In the
       latter case (L3), it is possible that no dynamic routing protocol is
       enabled. Tenant systems can be multi-homed into remote NVE using
       several interfaces (physical NICS or vNICS) with an IP address per
       interface either to the need for ARP, and
       hence addresses same nvo3 network or into different nvo3
       networks. When one of the issues links fails, the corresponding IP is not
       reachable but the other interfaces can still be used. When a tenant
       system is co-located with explosive ARP handling. Tunnel management

       A control plane protocol may an NVE, IP routing can be required relied upon to exchange tunnel state
       information. This may include setting up tunnels and/or providing
       tunnel state information.

       This applies
       handle routing over diverse links to both unicast TORs.

       External connectivity is handled by to or more nvo3 gateways. Each
       gateway is connected to a different domain (e.g. ISP) and multicast tunnels.

       For instance, it may be necessary runs BGP
       multi-homing. They serve as an access point to provide active/standby status
       information between NVEs, up/down status information,
       pruning/grafting information for multicast tunnels, etc.

    3.2. external networks
       such as VPNs or the Internet. When a connection to an upstream
       router is lost, the alternative connection is used and the failed
       route withdrawn.

    3.3. Service Overlay Topologies

       A number of service topologies may be used to optimize the service
       connectivity and to address NVE performance limitations.

       The topology described in Figure 3 suggests the use of a tunnel mesh
       between the NVEs where each tenant instance is one hop away from a
       service processing perspective. Partial mesh topologies and an NVE
       hierarchy may be used where certain NVEs may act as service transit

    4. Key aspects of overlay networks

       The intent of this section is to highlight specific issues that
       proposed overlay solutions need to address.

    4.1. Pros & Cons

       An overlay network is a layer of virtual network topology on top of
       the physical network.

       Overlay networks offer the following key advantages:

          o Unicast tunneling state management is handled at the edge of
            the network. Intermediate transport nodes are unaware of such
            state. Note that this is not the case when multicast is enabled
            in the core network.

          o Tunnels are used to aggregate traffic and hence offer the
            advantage of minimizing the amount of forwarding state required
            within the underlay network

          o Decoupling of the overlay addresses (MAC and IP) used by VMs
            from the underlay network. This offers a clear separation
            between addresses used within the overlay and the underlay
            networks and it enables the use of overlapping addresses spaces
            by Tenant End Systems

          o Support of a large number of virtual network identifiers

       Overlay networks also create several challenges:

          o Overlay networks have no controls of underlay networks and lack
            critical network information
               o Overlays typically probe the network to measure link
                 properties, such as available bandwidth or packet loss
                 rate. It is difficult to accurately evaluate network
                 properties. It might be preferable for the underlay
                 network to expose usage and performance information.

          o Miscommunication between overlay and underlay networks can lead
            to an inefficient usage of network resources.

          o Fairness of resource sharing and collaboration among end-nodes
            in overlay networks are two critical issues

          o When multiple overlays co-exist on top of a common underlay
            network, the lack of coordination between overlays can lead to
            performance issues.

          o Overlaid traffic may not traverse firewalls and NAT devices.

          o Multicast service scalability. Multicast support may be
            required in the overlay network to address for each tenant
            flood containment or efficient multicast handling.

          o Hash-based load balancing may not be optimal as the hash
            algorithm may not work well due to the limited number of
            combinations of tunnel source and destination addresses

    4.2. Overlay issues to consider

    4.2.1. Data plane vs Control plane driven

       In the case of an L2NVE, it is possible to dynamically learn MAC
       addresses against VAPs. It is also possible that such addresses be
       known and controlled via management or a control protocol for both
       L2NVEs and L3NVEs.

       Dynamic data plane learning implies that flooding of unknown
       destinations be supported and hence implies that broadcast and/or
       multicast be supported. Multicasting in the core network for dynamic
       learning may lead to significant scalability limitations. Specific
       forwarding rules must be enforced to prevent loops from happening.
       This can be achieved using a spanning tree, a shortest path tree, or
       a split-horizon mesh.

       It should be noted that the amount of state to be distributed is
       dependent upon network topology and the number of virtual machines.
       Different forms of caching can also be utilized to minimize state
       distribution between the various elements. The control plane should
       not require an NVE to maintain the locations of all the tenant
       systems whose VNs are not present on the NVE.

    4.2.2. Coordination between data plane and control plane

       For an L2 NVE, the NVE needs to be able to determine MAC addresses
       of the end systems present on a VAP (for instance, VAP. This can be achieved via
       dataplane learning may be relied upon for this purpose). or a control plane. For an L3 NVE, the NVE needs
       to be able to determine IP addresses of the end systems present on a

       In both cases, coordination with the NVE control protocol is needed
       such that when the NVE determines that the set of addresses behind a
       VAP has changed, it triggers the local NVE control plane to
       distribute this information to its peers.

    4.2.3. Handling Broadcast, Unknown Unicast and Multicast (BUM) traffic

       There are two techniques to support packet replication needed for
       broadcast, unknown unicast and multicast:

          o Ingress replication

          o Use of core multicast trees

       There is a bandwidth vs state trade-off between the two approaches.
       Depending upon the degree of replication required (i.e. the number
       of hosts per group) and the amount of multicast state to maintain,
       trading bandwidth for state is of consideration.

       When the number of hosts per group is large, the use of core
       multicast trees may be more appropriate. When the number of hosts is
       small (e.g. 2-3), ingress replication may not be an issue.

       Depending upon the size of the data center network and hence the
       number of (S,G) entries, but also the duration of multicast flows,
       the use of core multicast trees can be a challenge.

       When flows are well known, it is possible to pre-provision such
       multicast trees. However, it is often difficult to predict
       application flows ahead of time, and hence programming of (S,G)
       entries for short-lived flows could be impractical.

       A possible trade-off is to use in the core shared multicast trees as
       opposed to dedicated multicast trees.

    4.2.4. Path MTU

       When using overlay tunneling, an outer header is added to the
       original frame. This can cause the MTU of the path to the egress
       tunnel endpoint to be exceeded.

       In this section, we will only consider the case of an IP overlay.

       It is usually not desirable to rely on IP fragmentation for
       performance reasons. Ideally, the interface MTU as seen by a Tenant
       System is adjusted such that no fragmentation is needed. TCP will
       adjust its maximum segment size accordingly.

       It is possible for the MTU to be configured manually or to be
       discovered dynamically. Various Path MTU discovery techniques exist
       in order to determine the proper MTU size to use:

          o Classical ICMP-based MTU Path Discovery [RFC1191] [RFC1981]

                Tenant End Systems rely on ICMP messages to discover the MTU of
                 the end-to-end path to its destination. This method is not
                 always possible, such as when traversing middle boxes
                 (e.g. firewalls) which disable ICMP for security reasons

          o Extended MTU Path Discovery techniques such as defined in

       It is also possible to rely on the overlay layer to perform
       segmentation and reassembly operations without relying on the Tenant
       Systems to know about the end-to-end MTU. The assumption is that
       some hardware assist is available on the NVE node to perform such
       SAR operations. However, fragmentation by the overlay layer can lead
       to performance and congestion issues due to TCP dynamics and might
       require new congestion avoidance mechanisms from then underlay
       network [FLOYD].

       Finally, the underlay network may be designed in such a way that the
       MTU can accommodate the extra tunnel overhead.

    4.2.5. NVE location trade-offs

       In the case of DC traffic, traffic originated from a VM is native
       Ethernet traffic. This traffic can be switched by a local VM switch
       or ToR switch and then by a DC gateway. The NVE function can be
       embedded within any of these elements.

       There are several criteria to consider when deciding where the NVE
       processing boundary happens:

          o Processing and memory requirements

              o Datapath (e.g. lookups, filtering,

              o Control plane processing (e.g. routing, signaling, OAM)

          o FIB/RIB size

          o Multicast support

              o Routing protocols

              o Packet replication capability

          o Fragmentation support

          o QoS transparency

          o Resiliency

    4.2.6. Interaction between network overlays and underlays

       When multiple overlays co-exist on top of a common underlay network,
       this can cause some performance issues. These overlays have
       partially overlapping paths and nodes.

       Each overlay is selfish by nature in that it sends traffic so as to
       optimize its own performance without considering the impact on other
       overlays, unless the underlay tunnels are traffic engineered on a
       per overlay basis so as to avoid sharing underlay resources.

       Better visibility between overlays and underlays can be achieved by
       providing mechanisms to exchange information about:

          o Performance metrics (throughput, delay, loss, jitter)

          o Cost metrics

    5. Security Considerations

       As a framework document, no protocols are being defined and hence no
       specific security consideration are raised.

       The following security aspects shall be discussed in respective
       solutions documents:

       Traffic isolation between NVO3 domains is guaranteed by the use of
       per tenant FIB tables (VNIs).

       The creation of overlay networks and the tenant to overlay mapping
       function can introduce significant security risks if appropriate risks. When dynamic
       protocols are not used used, authentication should be supported. When a
       centralized controller is used, access to that controller should be
       restricted to authorized personnel. This can
       support mutual be achieved via login

       No other new security issues are introduced beyond those described
       already in the related L2VPN and L3VPN RFCs.

    6. IANA Considerations

       IANA does not need to take any action for this draft.

    7. References

    7.1. Normative References

       [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
                 Requirement Levels", BCP 14, RFC 2119, March 1997.

    7.2. Informative References

       [NVOPS] Narten, T. et al, "Problem Statement : Overlays for Network
                 Virtualization", draft-narten-nvo3-overlay-problem-
                 statement (work in progress)

       [OVCPREQ] Kreeger, L. et al, "Network Virtualization Overlay Control
                 Protocol Requirements", draft-kreeger-nvo3-overlay-cp
                 (work in progress)

       [FLOYD] Sally Floyd, Allyn Romanow, "Dynamics of TCP Traffic over
                 ATM Networks", IEEE JSAC, V. 13 N. 4, May 1995

       [RFC4364] Rosen, E. and Y. Rekhter, "BGP/MPLS IP Virtual Private
                 Networks (VPNs)", RFC 4364, February 2006.

       [RFC1191] Mogul, J. "Path MTU Discovery", RFC1191, November 1990

       [RFC1981] McCann, J. et al, "Path MTU Discovery for IPv6", RFC1981,
                 August 1996

       [RFC4821] Mathis, M. et al, "Packetization Layer Path MTU
                 Discovery", RFC4821, March 2007

    8. Acknowledgments

       In addition to the authors the following people have contributed to
       this document:

       Dimitrios Stiliadis, Rotem Salomonovitch, Alcatel-Lucent

       Lucy Yong, Huawei

       This document was prepared using

    Authors' Addresses

       Marc Lasserre

       Florin Balus
       777 E. Middlefield Road
       Mountain View, CA, USA 94043

       Thomas Morin
       France Telecom Orange

       Nabil Bitar
       40 Sylvan Road
       Waltham, MA 02145

       Yakov Rekhter