Internet Engineering Task Force                            Nabil Bitar
    Internet Draft                                                 Verizon
    Intended status: Informational
    Expires: January 2014                                    Marc Lasserre
                                                              Florin Balus
                                                              Thomas Morin
                                                     France Telecom Orange
                                                               Lizhong Jin
                                                         Bhumip Khasnabish
                                                              July 1, 2013
                           NVO3 Data Plane Requirements
    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 January 1, 2013.
    Copyright Notice
       Copyright (c) 2013 IETF Trust and the persons identified as the
       document authors. All rights reserved.
    Lasserre, et al.       Expires January 1, 2013                [Page 1]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       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 list of data
       plane requirements for Network Virtualization over L3 (NVO3) that
       have to be addressed in solutions documents.
    Table of Contents
       1. Introduction................................................3
          1.1. Conventions used in this document.......................3
          1.2. General terminology.....................................3
       2. Data Path Overview..........................................4
       3. Data Plane Requirements......................................5
          3.1. Virtual Access Points (VAPs)............................5
          3.2. Virtual Network Instance (VNI)..........................5
          3.2.1. L2 VNI...............................................5
          3.2.2. L3 VNI...............................................6
          3.3. Overlay Module.........................................7
          3.3.1. NVO3 overlay header...................................8
 Virtual Network Context Identification..............8
 Service QoS identifier..............................8
          3.3.2. Tunneling function....................................9
 LAG and ECMP.......................................10
 DiffServ and ECN marking...........................10
 Handling of BUM traffic............................11
          3.4. External NVO3 connectivity.............................11
          3.4.1. GW Types............................................12
 VPN and Internet GWs...............................12
 Inter-DC GW........................................12
 Intra-DC gateways..................................12
    Lasserre, et al.       Expires January 1, 2013                [Page 2]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
          3.4.2. Path optimality between NVEs and Gateways............12
 Triangular Routing Issues (Traffic Tromboning)......13
          3.5. Path MTU..............................................14
          3.6. Hierarchical NVE.......................................15
          3.7. NVE Multi-Homing Requirements..........................15
          3.8. OAM...................................................16
          3.9. Other considerations...................................16
          3.9.1. Data Plane Optimizations.............................16
          3.9.2. NVE location trade-offs..............................17
       4. Security Considerations.....................................17
       5. IANA Considerations........................................17
       6. References.................................................18
          6.1. Normative References...................................18
          6.2. Informative References.................................18
       7. Acknowledgments............................................19
    1. Introduction
    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
       The terminology defined in [NVO3-framework] is used throughout this
       document. Terminology specific to this memo is defined here and is
       introduced as needed in later sections.
       BUM: Broadcast, Unknown Unicast, Multicast traffic
       TS: Tenant System
    Lasserre, et al.       Expires January 1, 2013                [Page 3]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
    2. Data Path Overview
       The NVO3 framework [NVO3-framework] defines the generic NVE model
       depicted in Figure 1:
                          +------- L3 Network ------+
                          |                         |
                          |       Tunnel Overlay    |
             +------------+---------+       +---------+------------+
             | +----------+-------+ |       | +---------+--------+ |
             | |  Overlay Module  | |       | |  Overlay Module  | |
             | +---------+--------+ |       | +---------+--------+ |
             |           |VN context|       | VN context|          |
             |           |          |       |           |          |
             |  +-------+--------+  |       |  +--------+-------+  |
             |  | |VNI|  ... |VNI|  |       |  | |VNI|  ... |VNI|  |
        NVE1 |  +-+------------+-+  |       |  +-+-----------+--+  | NVE2
             |    |   VAPs     |    |       |    |    VAPs   |     |
             +----+------------+----+       +----+------------+----+
                 |            |                 |            |
                 |            |     Tenant      |            |
                 |            |   Service IF    |            |
                 Tenant Systems                 Tenant Systems
                  Figure 1 : Generic reference model for NV Edge
       When a frame is received by an ingress NVE from a Tenant System over
       a local VAP, it needs to be parsed in order to identify which
       virtual network instance it belongs to. The parsing function can
       examine various fields in the data frame (e.g., VLANID) and/or
       associated interface/port the frame came from.
       Once a corresponding VNI is identified, a lookup is performed to
       determine where the frame needs to be sent. This lookup can be based
       on any combinations of various fields in the data frame (e.g.,
       destination MAC addresses and/or destination IP addresses). Note
       that additional criteria such as 802.1p and/or DSCP markings might
       be used to select an appropriate tunnel or local VAP destination.
       Lookup tables can be populated using different techniques: data
       plane learning, management plane configuration, or a distributed
       control plane. Management and control planes are not in the scope of
    Lasserre, et al.       Expires January 1, 2013                [Page 4]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       this document. The data plane based solution is described in this
       document as it has implications on the data plane processing
       The result of this lookup yields the corresponding information
       needed to build the overlay header, as described in section 3.3.
       This information includes the destination L3 address of the egress
       NVE. Note that this lookup might yield a list of tunnels such as
       when ingress replication is used for BUM traffic.
       The overlay header MUST include a context identifier which the
       egress NVE will use to identify which VNI this frame belongs to.
       The egress NVE checks the context identifier and removes the
       encapsulation header and then forwards the original frame towards
       the appropriate recipient, usually a local VAP.
    3. Data Plane Requirements
    3.1. Virtual Access Points (VAPs)
       The NVE forwarding plane MUST support VAP identification through the
       following mechanisms:
       - Using the local interface on which the frames are received, where
          the local interface may be an internal, virtual port in a VSwitch
          or a physical port on the ToR
       - Using the local interface and some fields in the frame header,
          e.g. one or multiple VLANs or the source MAC
    3.2. Virtual Network Instance (VNI)
       VAPs are associated with a specific VNI at service instantiation
       A VNI identifies a per-tenant private context, i.e. per-tenant
       policies and a FIB table to allow overlapping address space between
       There are different VNI types differentiated by the virtual network
       service they provide to Tenant Systems. Network virtualization can
       be provided by L2 and/or L3 VNIs.
    3.2.1. L2 VNI
       An L2 VNI MUST provide an emulated Ethernet multipoint service as if
       Tenant Systems are interconnected by a bridge (but instead by using
    Lasserre, et al.       Expires January 1, 2013                [Page 5]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       a set of NVO3 tunnels). The emulated bridge MAY be 802.1Q enabled
       (allowing use of VLAN tags as a VAP). An L2 VNI provides per tenant
       virtual switching instance with MAC addressing isolation and L3
       tunneling. Loop avoidance capability MUST be provided.
       Forwarding table entries provide mapping information between tenant
       system MAC addresses and VAPs on directly connected VNIs and L3
       tunnel destination addresses over the overlay. Such entries MAY be
       populated by a control or management plane, or via data plane.
       In the absence of a management or control plane, data plane learning
       MUST be used to populate forwarding tables. As frames arrive from
       VAPs or from overlay tunnels, standard MAC learning procedures are
       used: The tenant system source MAC address is learned against the
       VAP or the NVO3 tunneling encapsulation source address on which the
       frame arrived. This implies that unknown unicast traffic be flooded
       i.e. broadcast.
       When flooding is required, either to deliver unknown unicast, or
       broadcast or multicast traffic, the NVE MUST either support ingress
       replication or multicast. In this latter case, the NVE MUST have one
       or more multicast trees that can be used by local VNIs for flooding
       to NVEs belonging to the same VN. For each VNI, there is one
       flooding tree, and a multicast tree may be dedicated per VNI or
       shared across VNIs. In such cases, multiple VNIs MAY share the same
       default flooding tree.  The flooding tree is equivalent with a
       multicast (*,G) construct where all the NVEs for which the
       corresponding VNI is instantiated are members. The multicast tree
       MAY be established automatically via routing and signaling or pre-
       When tenant multicast is supported, it SHOULD also be possible to
       select whether the NVE provides optimized multicast trees inside the
       VNI for individual tenant multicast groups or whether the default
       VNI flooding tree is used. If the former option is selected the VNI
       SHOULD be able to snoop IGMP/MLD messages in order to efficiently
       join/prune Tenant System from multicast trees.
    3.2.2. L3 VNI
       L3 VNIs MUST provide virtualized IP routing and forwarding. L3 VNIs
       MUST support per-tenant forwarding instance with IP addressing
       isolation and L3 tunneling for interconnecting instances of the same
       VNI on NVEs.
       In the case of L3 VNI, the inner TTL field MUST be decremented by
       (at least) 1 as if the NVO3 egress NVE was one (or more) hop(s)
    Lasserre, et al.       Expires January 1, 2013                [Page 6]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       away. The TTL field in the outer IP header MUST be set to a value
       appropriate for delivery of the encapsulated frame to the tunnel
       exit point. Thus, the default behavior MUST be the TTL pipe model
       where the overlay network looks like one hop to the sending NVE.
       Configuration of a "uniform" TTL model where the outer tunnel TTL is
       set equal to the inner TTL on ingress NVE and the inner TTL is set
       to the outer TTL value on egress MAY be supported.
       L2 and L3 VNIs can be deployed in isolation or in combination to
       optimize traffic flows per tenant across the overlay network. For
       example, an L2 VNI may be configured across a number of NVEs to
       offer L2 multi-point service connectivity while a L3 VNI can be co-
       located to offer local routing capabilities and gateway
       functionality. In addition, integrated routing and bridging per
       tenant MAY be supported on an NVE. An instantiation of such service
       may be realized by interconnecting an L2 VNI as access to an L3 VNI
       on the NVE.
       The L3 VNI does not require support for Broadcast and Unknown
       Unicast traffic. The L3 VNI MAY provide support for customer
       multicast groups. When multicast is supported, it SHOULD be possible
       to select whether the NVE provides optimized multicast trees inside
       the VNI for individual tenant multicast groups or whether a default
       VNI multicasting tree, where all the NVEs of the corresponding VNI
       are members, is used.
    3.3. Overlay Module
       The overlay module performs a number of functions related to NVO3
       header and tunnel processing.
       The following figure shows a generic NVO3 encapsulated frame:
                           |     Tenant Frame         |
                           |   NVO3 Overlay Header    |
                           |   Outer Underlay header  |
                           |  Outer Link layer header |
                        Figure 2 : NVO3 encapsulated frame
    Lasserre, et al.       Expires January 1, 2013                [Page 7]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
            . Tenant frame: Ethernet or IP based upon the VNI type
            . NVO3 overlay header: Header containing VNI context information
              and other optional fields that can be used for processing
              this packet.
            . Outer underlay header: Can be either IP or MPLS
            . Outer link layer header: Header specific to the physical
              transmission link used
    3.3.1. NVO3 overlay header
       An NVO3 overlay header MUST be included after the underlay tunnel
       header when forwarding tenant traffic. Note that this information
       can be carried within existing protocol headers (when overloading of
       specific fields is possible) or within a separate header. Virtual Network Context Identification
       The overlay encapsulation header MUST contain a field 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 identifier).
       It SHOULD be aligned on a 32-bit boundary so as to make it
       efficiently processable by the data path. It MUST be distributable
       by a control-plane or configured via a management plane.
       In the case of a global identifier, this field MUST be large enough
       to scale to 100's of thousands of virtual networks. Note that there
       is no such constraint when using a local identifier. Service QoS identifier
       Traffic flows originating from different applications could rely on
       differentiated forwarding treatment to meet end-to-end availability
       and performance objectives. Such applications may span across one or
       more overlay networks. To enable such treatment, support for
    Lasserre, et al.       Expires January 1, 2013                [Page 8]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       multiple Classes of Service across or between overlay networks MAY
       be required.
       To effectively enforce CoS across or between overlay networks, NVEs
       MAY be able to map CoS markings between networking layers, e.g.,
       Tenant Systems, Overlays, and/or Underlay, enabling each networking
       layer to independently enforce its own CoS policies. For example:
       - TS (e.g. VM) CoS
            o  Tenant CoS policies MAY be defined by Tenant administrators
            o  QoS fields (e.g. IP DSCP and/or Ethernet 802.1p) in the
               tenant frame are used to indicate application level CoS
       - NVE CoS
            o  NVE MAY classify packets based on Tenant CoS markings or
               other mechanisms (eg. DPI) to identify the proper service CoS
               to be applied across the overlay network
            o  NVE service CoS levels are normalized to a common set (for
               example 8 levels) across multiple tenants; NVE uses per
               tenant policies to map Tenant CoS to the normalized service
               CoS fields in the NVO3 header
       - Underlay CoS
            o  The underlay/core network MAY use a different CoS set (for
               example 4 levels) than the NVE CoS as the core devices MAY
               have different QoS capabilities compared with NVEs.
            o  The Underlay CoS MAY also change as the NVO3 tunnels pass
               between different domains.
       Support for NVE Service CoS MAY be provided through a QoS field,
       inside the NVO3 overlay header. Examples of service CoS provided
       part of the service tag are 802.1p and DE bits in the VLAN and PBB
       ISID tags and MPLS TC bits in the VPN labels.
    3.3.2. Tunneling function
       This section describes the underlay tunneling requirements. From an
       encapsulation perspective, IPv4 or IPv6 MUST be supported, both IPv4
       and IPv6 SHOULD be supported, MPLS tunneling MAY be supported.
    Lasserre, et al.       Expires January 1, 2013                [Page 9]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
     LAG and ECMP
       For performance reasons, multipath over LAG and ECMP paths SHOULD be
       LAG (Link Aggregation Group) [IEEE 802.1AX-2008] and ECMP (Equal
       Cost Multi Path) are commonly used techniques to perform load-
       balancing of microflows over a set of a parallel links either at
       Layer-2 (LAG) or Layer-3 (ECMP). Existing deployed hardware
       implementations of LAG and ECMP uses a hash of various fields in the
       encapsulation (outermost) header(s) (e.g. source and destination MAC
       addresses for non-IP traffic, source and destination IP addresses,
       L4 protocol, L4 source and destination port numbers, etc).
       Furthermore, hardware deployed for the underlay network(s) will be
       most often unaware of the carried, innermost L2 frames or L3 packets
       transmitted by the TS. Thus, in order to perform fine-grained load-
       balancing over LAG and ECMP paths in the underlying network, the
       encapsulation MUST result in sufficient entropy to exercise all
       paths through several LAG/ECMP hops. The entropy information MAY be
       inferred from the NVO3 overlay header or underlay header. If the
       overlay protocol does not support the necessary entropy information
       or the switches/routers in the underlay do not support parsing of
       the additional entropy information in the overlay header, underlay
       switches and routers should be programmable, i.e. select the
       appropriate fields in the underlay header for hash calculation based
       on the type of overlay header.
       All packets that belong to a specific flow MUST follow the same path
       in order to prevent packet re-ordering. This is typically achieved
       by ensuring that the fields used for hashing are identical for a
       given flow.
       All paths available to the overlay network SHOULD be used
       efficiently. Different flows SHOULD be distributed as evenly as
       possible across multiple underlay network paths. For instance, this
       can be achieved by ensuring that some fields used for hashing are
       randomly generated. DiffServ and ECN marking
       When traffic is encapsulated in a tunnel header, there are numerous
       options as to how the Diffserv Code-Point (DSCP) and Explicit
       Congestion Notification (ECN) markings are set in the outer header
       and propagated to the inner header on decapsulation.
       [RFC2983] defines two modes for mapping the DSCP markings from inner
       to outer headers and vice versa.  The Uniform model copies the inner
    Lasserre, et al.       Expires January 1, 2013               [Page 10]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       DSCP marking to the outer header on tunnel ingress, and copies that
       outer header value back to the inner header at tunnel egress.  The
       Pipe model sets the DSCP value to some value based on local policy
       at ingress and does not modify the inner header on egress.  Both
       models SHOULD be supported.
       ECN marking MUST be performed according to [RFC6040] which describes
       the correct ECN behavior for IP tunnels. Handling of BUM traffic
       NVO3 data plane support for either ingress replication or point-to-
       multipoint tunnels is required to send traffic destined to multiple
       locations on a per-VNI basis (e.g. L2/L3 multicast traffic, L2
       broadcast and unknown unicast traffic). It is possible that both
       methods be used simultaneously.
       There is a bandwidth vs state trade-off between the two approaches.
       User-definable knobs MUST be provided to select which method(s) gets
       used based upon the amount of replication required (i.e. the number
       of hosts per group), the amount of multicast state to maintain, the
       duration of multicast flows and the scalability of multicast
       When ingress replication is used, NVEs MUST track for each VNI the
       related tunnel endpoints to which it needs to replicate the frame.
       For point-to-multipoint tunnels, the bandwidth efficiency is
       increased at the cost of more state in the Core nodes. The ability
       to auto-discover or pre-provision the mapping between VNI multicast
       trees to related tunnel endpoints at the NVE and/or throughout the
       core SHOULD be supported.
    3.4. External NVO3 connectivity
       NVO3 services MUST interoperate with current VPN and Internet
       services. This may happen inside one DC during a migration phase or
       as NVO3 services are delivered to the outside world via Internet or
       VPN gateways.
       Moreover the compute and storage services delivered by a NVO3 domain
       may span multiple DCs requiring Inter-DC connectivity. From a DC
       perspective a set of gateway devices are required in all of these
       cases albeit with different functionalities influenced by the
       overlay type across the WAN, the service type and the DC network
       technologies used at each DC site.
    Lasserre, et al.       Expires January 1, 2013               [Page 11]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       A GW handling the connectivity between NVO3 and external domains
       represents a single point of failure that may affect multiple tenant
       services. Redundancy between NVO3 and external domains MUST be
    3.4.1. GW Types VPN and Internet GWs
       Tenant sites may be already interconnected using one of the existing
       VPN services and technologies (VPLS or IP VPN). If a new NVO3
       encapsulation is used, a VPN GW is required to forward traffic
       between NVO3 and VPN domains. Translation of encapsulations MAY be
       required. Internet connected Tenants require translation from NVO3
       encapsulation to IP in the NVO3 gateway. The translation function
       SHOULD minimize provisioning touches. Inter-DC GW
       Inter-DC connectivity MAY be required to provide support for
       features like disaster prevention or compute load re-distribution.
       This MAY be provided via a set of gateways interconnected through a
       WAN. This type of connectivity MAY be provided either through
       extension of the NVO3 tunneling domain or via VPN GWs. Intra-DC gateways
       Even within one DC there may be End Devices that do not support NVO3
       encapsulation, for example bare metal servers, hardware appliances
       and storage. A gateway device, e.g. a ToR, is required to translate
       the NVO3 to Ethernet VLAN encapsulation.
    3.4.2. Path optimality between NVEs and Gateways
       Within the NVO3 overlay, a default assumption is that NVO3 traffic
       will be equally load-balanced across the underlying network
       consisting of LAG and/or ECMP paths. This assumption is valid only
       as long as: a) all traffic is load-balanced equally among each of
       the component-links and paths; and, b) each of the component-
       links/paths is of identical capacity. During the course of normal
       operation of the underlying network, it is possible that one, or
       more, of the component-links/paths of a LAG may be taken out-of-
       service in order to be repaired, e.g.: due to hardware failure of
       cabling, optics, etc. In such cases, the administrator should
       configure the underlying network such that an entire LAG bundle in
       the underlying network will be reported as operationally down if
       there is a failure of any single component-link member of the LAG
    Lasserre, et al.       Expires January 1, 2013               [Page 12]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       bundle, (e.g.: N = M configuration of the LAG bundle), and, thus,
       they know that traffic will be carried sufficiently by alternate,
       available (potentially ECMP) paths in the underlying network. This
       is a likely an adequate assumption for Intra-DC traffic where
       presumably the costs for additional, protection capacity along
       alternate paths is not cost-prohibitive. Thus, there are likely no
       additional requirements on NVO3 solutions to accommodate this type
       of underlying network configuration and administration.
       There is a similar case with ECMP, used Intra-DC, where failure of a
       single component-path of an ECMP group would result in traffic
       shifting onto the surviving members of the ECMP group.
       Unfortunately, there are no automatic recovery methods in IP routing
       protocols to detect a simultaneous failure of more than one
       component-path in a ECMP group, operationally disable the entire
       ECMP group and allow traffic to shift onto alternative paths. This
       problem is attributable to the underlying network and, thus, out-of-
       scope of any NVO3 solutions.
       On the other hand, for Inter-DC and DC to External Network cases
       that use a WAN, the costs of the underlying network and/or service
       (e.g.: IPVPN service) are more expensive; therefore, there is a
       requirement on administrators to both: a) ensure high availability
       (active-backup failover or active-active load-balancing); and, b)
       maintaining substantial utilization of the WAN transport capacity at
       nearly all times, particularly in the case of active-active load-
       balancing. With respect to the dataplane requirements of NVO3
       solutions, in the case of active-backup fail-over, all of the
       ingress NVE's MUST dynamically adapt to the failure of an active NVE
       GW when the backup NVE GW announces itself into the NVO3 overlay
       immediately following a failure of the previously active NVE GW and
       update their forwarding tables accordingly, (e.g.: perhaps through
       dataplane learning and/or translation of a gratuitous ARP, IPv6
       Router Advertisement, etc.) Note that active-backup fail-over could
       be used to accomplish a crude form of load-balancing by, for
       example, manually configuring each tenant to use a different NVE GW,
       in a round-robin fashion. On the other hand, with respect to active-
       active load-balancing across physically separate NVE GW's (e.g.:
       two, separate chassis) an NVO3 solution SHOULD support forwarding
       tables that can simultaneously map a single egress NVE to more than
       one NVO3 tunnels. The granularity of such mappings, in both active-
       backup and active-active, MUST be unique to each tenant. Triangular Routing Issues (Traffic Tromboning)
       L2/ELAN over NVO3 service may span multiple racks distributed across
       different DC regions. Multiple ELANs belonging to one tenant may be
    Lasserre, et al.       Expires January 1, 2013               [Page 13]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
       interconnected or connected to the outside world through multiple
       Router/VRF gateways distributed throughout the DC regions. In this
       scenario, without aid from an NVO3 or other type of solution,
       traffic from an ingress NVE destined to External gateways will take
       a non-optimal path that will result in higher latency and costs,
       (since it is using more expensive resources of a WAN). In the case
       of traffic from an IP/MPLS network destined toward the entrance to
       an NVO3 overlay, well-known IP routing techniques MAY be used to
       optimize traffic into the NVO3 overlay, (at the expense of
       additional routes in the IP/MPLS network). In summary, these issues
       are well known as triangular routing.
       Procedures for gateway selection to avoid triangular routing issues
       SHOULD be provided. The details of such procedures are, most likely,
       part of the NVO3 Management and/or Control Plane requirements and,
       thus, out of scope of this document. However, a key requirement on
       the dataplane of any NVO3 solution to avoid triangular routing is
       stated above, in Section 3.4.2, with respect to active-active load-
       balancing. More specifically, an NVO3 solution SHOULD support
       forwarding tables that can simultaneously map a single egress NVE to
       more than one NVO3 tunnels. The expectation is that, through the
       Control and/or Management Planes, this mapping information MAY be
       dynamically manipulated to, for example, provide the closest
       geographic and/or topological exit point (egress NVE) for each
       ingress NVE.
    3.5. Path MTU
       The tunnel overlay header can cause the MTU of the path to the
       egress tunnel endpoint to be exceeded.
       IP fragmentation SHOULD be avoided for performance reasons.
       The interface MTU as seen by a Tenant System SHOULD be adjusted such
       that no fragmentation is needed. This can be achieved by
       configuration or be discovered dynamically.
       Either of the following options MUST be supported:
          o Classical ICMP-based MTU Path Discovery [RFC1191] [RFC1981] or
            Extended MTU Path Discovery techniques such as defined in
          o Segmentation and reassembly support from the overlay layer
            operations without relying on the Tenant Systems to know about
            the end-to-end MTU
    Lasserre, et al.       Expires January 1, 2013               [Page 14]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
          o The underlay network MAY be designed in such a way that the MTU
            can accommodate the extra tunnel overhead.
    3.6. Hierarchical NVE
       It might be desirable to support the concept of hierarchical NVEs,
       such as spoke NVEs and hub NVEs, in order to address possible NVE
       performance limitations and service connectivity optimizations.
       For instance, spoke NVE functionality MAY be used when processing
       capabilities are limited. A hub NVE would provide additional data
       processing capabilities such as packet replication.
       NVEs can be either connected in an any-to-any or hub and spoke
       topology on a per VNI basis.
    3.7. NVE Multi-Homing Requirements
       Multi-homing techniques SHOULD 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
       Tenant systems can either be L2 or L3 nodes. In the former case
       (L2), techniques such as LAG or STP for instance MAY 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 same nvo3 network or into different nvo3
       networks. When one of the links fails, the corresponding IP is not
       reachable but the other interfaces can still be used. When a tenant
       system is co-located with an NVE, IP routing can be relied upon to
       handle routing over diverse links to TORs.
       External connectivity MAY be handled by two or more nvo3 gateways.
       Each gateway is connected to a different domain (e.g. ISP) and runs
       BGP multi-homing. They serve as an access point to 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.
    Lasserre, et al.       Expires January 1, 2013               [Page 15]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
    3.8. OAM
       NVE MAY be able to originate/terminate OAM messages for connectivity
       verification, performance monitoring, statistic gathering and fault
       isolation. Depending on configuration, NVEs SHOULD be able to
       process or transparently tunnel OAM messages, as well as supporting
       alarm propagation capabilities.
       Given the critical requirement to load-balance NVO3 encapsulated
       packets over LAG and ECMP paths, it will be equally critical to
       ensure existing and/or new OAM tools allow NVE administrators to
       proactively and/or reactively monitor the health of various
       component-links that comprise both LAG and ECMP paths carrying NVO3
       encapsulated packets. For example, it will be important that such
       OAM tools allow NVE administrators to reveal the set of underlying
       network hops (topology) in order that the underlying network
       administrators can use this information to quickly perform fault
       isolation and restore the underlying network.
       The NVE MUST provide the ability to reveal the set of ECMP and/or
       LAG paths used by NVO3 encapsulated packets in the underlying
       network from an ingress NVE to egress NVE. The NVE MUST provide the
       ability to provide a "ping"-like functionality that can be used to
       determine the health (liveness) of remote NVE's or their VNI's. The
       NVE SHOULD provide a "ping"-like functionality to more expeditiously
       aid in troubleshooting performance problems, i.e.: blackholing or
       other types of congestion occurring in the underlying network, for
       NVO3 encapsulated packets carried over LAG and/or ECMP paths.
    3.9. Other considerations
    3.9.1. Data Plane Optimizations
       Data plane forwarding and encapsulation choices SHOULD consider the
       limitation of possible NVE implementations, specifically in software
       based implementations (e.g.  servers running VSwitches)
       NVE SHOULD provide efficient processing of traffic. For instance,
       packet alignment, the use of offsets to minimize header parsing,
       padding techniques SHOULD be considered when designing NVO3
       encapsulation types.
       The NV03 encapsulation/decapsulation processing in software-based
       NVEs SHOULD make use of hardware assist provided by NICs in order to
       speed up packet processing.
    Lasserre, et al.       Expires January 1, 2013               [Page 16]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
    3.9.2. 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.
       The NVE function can be supported in various DC network elements
       such as a VM, VM switch, ToR switch or DC GW.
       The following criteria SHOULD be considered 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. Security Considerations
       This requirements document does not raise in itself any specific
       security issues.
    5. IANA Considerations
       IANA does not need to take any action for this draft.
    Lasserre, et al.       Expires January 1, 2013               [Page 17]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
    6. References
    6.1. Normative References
       [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
                 Requirement Levels", BCP 14, RFC 2119, March 1997.
    6.2. Informative References
       [NVOPS] Narten, T. et al, "Problem Statement: Overlays for Network
                 Virtualization", draft-narten-nvo3-overlay-problem-
                 statement (work in progress)
       [NVO3-framework] Lasserre, M. et al, "Framework for DC Network
                 Virtualization", draft-lasserre-nvo3-framework (work in
       [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
       [RFC2983] Black, D. "Diffserv and tunnels", RFC2983, Cotober 2000
       [RFC6040] Briscoe, B. "Tunnelling of Explicit Congestion
                 Notification", RFC6040, November 2010
       [RFC6438] Carpenter, B. et al, "Using the IPv6 Flow Label for Equal
                 Cost Multipath Routing and Link Aggregation in Tunnels",
                 RFC6438, November 2011
       [RFC6391] Bryant, S. et al, "Flow-Aware Transport of Pseudowires
                 over an MPLS Packet Switched Network", RFC6391, November
    Lasserre, et al.       Expires January 1, 2013               [Page 18]

    Internet-Draft        NVO3 Data Plane Requirements        July 2013
    7. Acknowledgments
       In addition to the authors the following people have contributed to
       this document:
       Shane Amante, Level3
       Dimitrios Stiliadis, Rotem Salomonovitch, Alcatel-Lucent
       Larry Kreeger, Cisco
       This document was prepared using
    Authors' Addresses
       Nabil Bitar
       40 Sylvan Road
       Waltham, MA 02145
       Marc Lasserre
       Florin Balus
       777 E. Middlefield Road
       Mountain View, CA, USA 94043
       Thomas Morin
       France Telecom Orange
       Lizhong Jin
       Email :
       Bhumip Khasnabish
       Email :
    Lasserre, et al.       Expires January 1, 2013               [Page 19]