Internet Engineering Task Force                               L. Kreeger
Internet-Draft                                                     Cisco
Intended status: Informational                                   D. Dutt
Expires: January 18, 2013                               Cumulus Networks
                                                               T. Narten
                                                                D. Black
                                                            M. Sridharan
                                                           July 17, 2012

      Network Virtualization Overlay Control Protocol Requirements


   The document Overlays for Network Virtualization problem statement
   document discusses the needs for network virtualization using overlay
   networks in highly virtualized data centers.  The problem statement
   outlines a need for control protocols to facilitate running these
   overlay networks.  This document outlines the high level requirements
   to be fulfilled by the control protocols.

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 18, 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

Kreeger, et al.         Expires January 18, 2013                [Page 1]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   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.

Table of Contents

   1.  Introduction . . . . . . . . . . . . . . . . . . . . . . . . .  3
   2.  Terminology  . . . . . . . . . . . . . . . . . . . . . . . . .  3
   3.  Control Plane Protocol Functionality . . . . . . . . . . . . .  3
     3.1.  Inner to Outer Address Mapping . . . . . . . . . . . . . .  6
     3.2.  Underlying Network Multi-Destination Delivery
           Address(es)  . . . . . . . . . . . . . . . . . . . . . . .  6
     3.3.  VN Connect/Disconnect Notification . . . . . . . . . . . .  7
     3.4.  VN Name to VN-ID Mapping . . . . . . . . . . . . . . . . .  8
   4.  Control Plane Characteristics  . . . . . . . . . . . . . . . .  8
   5.  Security Considerations  . . . . . . . . . . . . . . . . . . . 10
   6.  Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 11
   7.  Informative References . . . . . . . . . . . . . . . . . . . . 11
   Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 11

Kreeger, et al.         Expires January 18, 2013                [Page 2]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

1.  Introduction

   "Problem Statement: Overlays for Network Virtualization"
   [I-D.narten-nvo3-overlay-problem-statement] discusses the needs for
   network virtualization using overlay networks in highly virtualized
   data centers and provides a general motivation for building such
   networks.  "Framework for DC Network Virtualization"
   [I-D.lasserre-nvo3-framework] provides a framework for discussing
   overlay networks generally and the various components that must work
   together in building such systems.  The reader is assumed to be
   familiar with both documents.

2.  Terminology

   This document uses the same terminology as found in
   [I-D.lasserre-nvo3-framework].  This section defines additional
   terminology used by this document.

   Network Service Appliance:  A stand-alone physical End Device or a
      virtual End Device that provides a network service, such as a
      firewall, load balancer, etc.  Such appliances may embed Network
      Virtualization Edge (NVE) functionality within them in order to
      more efficiently operate as part of a virtualized network.

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

   VN Name:  A globally unique name for a VN.  The VN Name is not
      carried in data packets originating from TES, but must be mapped
      into an appropriate VN-ID for a particular encapsulating
      technology.  Using VN Names rather than VN-IDs to identify VNs in
      configuration files and control protocols increases the
      portability of a VDC and its associated VNs when moving among
      different administrative domains (e.g. switching to a different
      cloud service provider).

3.  Control Plane Protocol Functionality

   The problem statement [I-D.narten-nvo3-overlay-problem-statement],
   discusses the needs for a control plane protocol (or protocols) to
   populate each NVE with the state needed to peform its functions.

   Note that an NVE may provide overlay encapsulation/decapsulation
   packet forwarding services to TESs that are co-resident within the

Kreeger, et al.         Expires January 18, 2013                [Page 3]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   same End Device (e.g. when the NVE is embedded within a hypervisor or
   a Network Service Appliance), or to TES that are externally connected
   to the NVE (e.g. a physical Network Service Appliance connected to an
   access switch containing the NVE).

   The following figures show examples of scenarios in which the NVE is
   co-resident within the same End Device as the TES connected to a
   given VN, and when the NVE is externally located within the access

   | +--+   +-------+---+  |
   | |VM|---|       |   |  |
   | +--+   |Virtual|NVE|----- Underlying
   | +--+   |Switch |   |  |    Network
   | |VM|---|       |   |  |
   | +--+   +-------+---+  |

   Hypervisor with an Embedded NVE.

                                 Figure 1

        Hypervisor             Access Switch
   +------------------+       +-----+-------+
   | +--+   +-------+ |       |     |       |
   | |VM|---|       | | VLAN  |     |       |
   | +--+   |Virtual|---------+ NVE |       +--- Underlying
   | +--+   |Switch | | Trunk |     |       |    Network
   | |VM|---|       | |       |     |       |
   | +--+   +-------+ |       |     |       |
   +------------------+       +-----+-------+

   Hypervisor with an External NVE.

                                 Figure 2

Kreeger, et al.         Expires January 18, 2013                [Page 4]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

    Network Service Appliance
   | +------------+   +-----+  |
   | |Net Service |---|     |  |
   | |Instance    |   |     |  |
   | +------------+   | NVE |------ Underlying
   | +------------+   |     |  |    Network
   | |Net Service |---|     |  |
   | |Instance    |   |     |  |
   | +------------+   +-----+  |

   Network Service Appliance (physical or virtual) with an Embedded NVE.

                                 Figure 3

    Network Service Appliance         Access Switch
   +--------------------------+      +-----+-------+
   | +------------+    |\     |      |     |       |
   | |Net Service |----| \    |      |     |       |
   | |Instance    |    |  \   | VLAN |     |       |
   | +------------+    |   |---------+ NVE |       +--- Underlying
   | +------------+    |   |  | Trunk|     |       |    Network
   | |Net Service |----|  /   |      |     |       |
   | |Instance    |    | /    |      |     |       |
   | +------------+    |/     |      |     |       |
   +--------------------------+      +-----+-------+

   Physical Network Service Appliance with an External NVE.

                                 Figure 4

   In the examples above where the NVE functionality is located in the
   physical access switch, the physical VLAN Trunk connecting the
   Hypervisor or Network Services Appliance to the external NVE only
   needs to carry locally significant (e.g. link local) VLAN tag values.
   These tags are only used to differentiate two different VNs as
   packets cross the wire to the external NVE.  When the NVE receives
   packets, it uses the VLAN tag to identify the VN the TES belongs to,
   strips the tag, and adds the appropriate overlay encapsulation for
   that VN.

   Given the above, a control plane protocol is necessary to provide an
   NVE with the information it needs to maintain its own internal state
   necessary to carry out its forwarding functions as explained in

Kreeger, et al.         Expires January 18, 2013                [Page 5]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   detail below.

   1.  An NVE maintains a per-VN table of mappings from TES (inner)
       addresses to Underlying Network (outer) addresses of remote NVEs.

   2.  An NVE maintains per-VN state for delivering multicast and
       broadcast packets to other TES.  Such state could include a list
       of multicast addresses and/or unicast addresses on the Underlying
       Network for the NVEs associated with a particular VN.

   3.  End Devices (such as a Hypervisor or Network Service Appliance)
       utilizing an external NVE need to "attach to" and "detach from"
       an NVE.  Specifically, they will need a protocol that runs across
       the L2 link between the two devices that identifies the TES and
       VN Name for which the NVE is providing service.  In addition,
       such a protocol will identify a locally significant tag (e.g., an
       802.1Q VLAN tag) that can be used to identify the data frames
       that flow between the TES and VN.

   4.  An NVE needs a mapping from each unique VN name to the VN-ID
       value used within encapsulated data packets within the
       administrative domain that the VN is instantiated.

3.1.  Inner to Outer Address Mapping

   When presented with a data packet to forward to an End System within
   a VN, the NVE needs to know the mapping of the TES destination
   (inner) address to the (outer) address on the Underlying Network of
   the remote NVE which can deliver the packet to the destination TES.

   A protocol is needed to provide this inner to outer mapping to each
   NVE that requires it and keep the mapping updated in a timely manner.
   Timely updates are important for maintaining connectivity between TES
   when one TES is a VM

   Note that one technique that could be used to create this mapping
   without the need for a control protocol is via data plane learning;
   However, the learning approach requires packets to be flooded to all
   NVEs participating in the VN when no mapping exists.  One goal of
   using a control protocol is to eliminate this flooding.

3.2.  Underlying Network Multi-Destination Delivery Address(es)

   Each NVE needs a way to deliver multi-destination packets (i.e.
   broadcast/multicast) within a given VN to each remote NVE which has a
   destination TES for these packets.  Three possible ways of
   accomplishing this:

Kreeger, et al.         Expires January 18, 2013                [Page 6]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   o  Use the multicast capabilities of the Underlying Network.

   o  Have each NVE replicate the packets and send a copy across the
      Underlying Network to each remote NVE currently participating in
      the VN.

   o  Use one or more distribution servers which replicates the packets
      on the behalf of the NVEs.

   Whichever method is used, a protocol is needed to provide on a per VN
   basis, one or more multicast address (assuming the Underlying Network
   supports multicast), and/or one or more unicast addresses of either
   the remote NVEs which are not multicast reachable, or of one or more
   distribution servers for the VN.

   The protocol must also keep the list of addresses up to date in a
   timely manner if the set of NVEs for a given VN changes over time.
   For example, the set of NVEs for a VN could change as VMs power on/
   off or migrate to different hypervisors.

3.3.  VN Connect/Disconnect Notification

   As the previous figures illustrated, NVEs may be embedded within an
   End Device (such as a Hypervisor or Network Service Appliance), or
   within an external networking device (e.g. an access switch).  Using
   an external network device as the NVE can provide an offload of the
   encapsulation / decapsulation function and the protocol overheads
   which may provide performance improvements and/or resource savings to
   the client End Device making use of the external NVE.

   When an NVE is external, a protocol is needed between a client End
   Device making use of the external NVE and the NVE itself in order to
   make the NVE aware of the changing VN membership requirements of the
   End Device.  A key driver for using a protocol rather than using
   static configuration of the exernal NVE is because the VN
   connectivity requirements can change frequently as VMs are brought
   up, moved and brought down on various hypervisors throughout the data

   The NVE must be notified when an End Device requires connection to a
   particular VN and when it no longer requires connection.  This
   protocol should also provide the inner TES addresses within the VN
   that the End Device contains (e.g. the virtual MAC address of a VMs
   virtual NIC) to the external NVE.  In addition, the external NVE must
   provide a local tag value for each connected VN to the End Device to
   use for exchange of packets between the End Device to the NVE (e.g. a
   locally significant 802.1Q tag value).

Kreeger, et al.         Expires January 18, 2013                [Page 7]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   The Identification of the VN in this protocol should preferably be
   made using a globally unique VN Name.  A globally unique VN Name
   facilitates portability of a Tenant's Virtual Data Center.  When a VN
   within a VDC is instantiated within a particular administrative
   domain, it can be allocated a VN-ID which only the NVE needs to use.
   An End Device that is making use of an offloaded NVE only needs to
   communicate the VN Name to the NVE, and get back a locally
   significant tag value.  Ideally the VN Name should be compact as well
   unique to minimize protocol overhead.  Using a Universally Unique
   Identifier (UUID) as discussed in RFC 4122, would work well because
   it is both compact and a fixed size and can be generated locally with
   a high likelihood of being unique.

3.4.  VN Name to VN-ID Mapping

   Once an NVE (embedded or external) receives a VN connect indication
   with a specified VN Name, the NVE must find the VN-ID value to
   encapsulate packets with.  The NVE serving that hypervisor needs a
   way to get a VN-ID allocated or receive the already allocated VN-ID
   for a given VN Name.  A protocol for an NVE to get this mapping may
   be a useful function.

4.  Control Plane Characteristics

   NVEs are expected to be implemented within hypervisors or access
   switches, or even within a Network Service Appliance.  Any resources
   used by these protocols (e.g. processing or memory) takes away
   resources that could be better used by these devices to perform their
   intended functions (e.g. providing resoures for hosted VMs).

   A large scale data center may contain hundreds of thousands of these
   NVEs (which may be several independent implementations); Therefore,
   any savings in per-NVE resources can be multiplied hundreds of
   thousands of times.

   Given this, the control plane protocol(s) implemented by NVEs to
   provide the functionality discussed above should have the below

   1.   Minimize the amount of state needed to be stored on each NVE.
        The NVE should only be required to cache state that it is
        actively using, and be able to discard any cached state when it
        is no longer required.  For example, an NVE should only need to
        maintain an inner-to-outer address mapping for destinations to
        which it is actively sending traffic as opposed to maintaining
        mappings for all possible destinations.

Kreeger, et al.         Expires January 18, 2013                [Page 8]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   2.   Fast acquisition of needed state.  For example, when a TES emits
        a packet destined to an inner address that the NVE does not have
        a mapping for, the NVE should be able to acquire the needed
        mapping quickly.

   3.   Fast detection/update of stale cached state information.  This
        only applies if the cached state is actually being used.  For
        example, when a VM moves such that it is connected to a
        different NVE, the inner to outer mapping for this VM's address
        that is cached on other NVEs must be updated in a timely manner
        (if they are actively in use).  If the update is not timely, the
        NVEs will forward data to the wrong NVE until it is updated.

   4.   Minimize processing overhead.  This means that an NVE should
        only be required to perform protocol processing directly related
        to maintaining state for the TESs it is actively communicating
        with.  This requirement is for the NVE functionality only.  The
        network node that contains the NVE may be involved in other
        functionality for the underlying network that maintains
        connectivity that the NVE is not actively using (e.g., routing
        and multicast distribution protocols for the underlying

   5.   Highly scalable.  This means scaling to hundreds of thousands of
        NVEs and several million VNs within a single administrative
        domain.  As the number of NVEs and/or VNs within a data center
        grows, the protocol overhead at any one NVE should not increase

   6.   Minimize the complexity of the implementation.  This argues for
        using the least number of protocols to achieve all the
        functionality listed above.  Ideally a single protocol should be
        able to be used.  The less complex the protocol is on the NVE,
        the more likely interoperable implementations will be created in
        a timely manner.

   7.   Extensible.  The protocol should easily accommodate extension to
        meet related future requirements.  For example, access control
        or QoS policies, or new address families for either inner or
        outer addresses should be easy to add while maintaining
        interoperability with NVEs running older versions.

   8.   Simple protocol configuration.  A minimal amount of
        configuration should be required for a new NVE to be
        provisioned.  Existing NVEs should not require any configuration
        changes when a new NVE is provisioned.  Ideally NVEs should be
        able to auto configure themselves.

Kreeger, et al.         Expires January 18, 2013                [Page 9]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   9.   Do not rely on IP Multicast in the Underlying Network.  Many
        data centers do not have IP multicast routing enabled.  If the
        Underlying Network is an IP network, the protocol should allow
        for, but not require the presence of IP multicast services
        within the data center.

   10.  Flexible mapping sources.  Inner to outer address mappings
        should be able to be created by either the NVEs themselves or
        other third party entities (e.g. data center management or
        orchestration systems).  The protocol should allow for mappings
        created by an NVE to be automatically removed from all other
        NVEs if it fails or is brought down unexpectedly.

   11.  Secure.  See the Security Considerations section below.

5.  Security Considerations

   Editor's Note: This is an initial start on the security
   considerations section; it will need to be expanded, and suggestions
   for material to add are welcome.

   The protocol(s) should protect the integrity of the mapping against
   both off-path and on-path attacks.  It should authenticate the
   systems that are creating mappings, and rely on light weight security
   mechanisms to minimize the impact on scalability and allow for simple

   Use of an overlay exposes virtual networks to attacks on the
   underlying network beyond attacks on the control protocol that is the
   subject of this draft.  In addition to the directly applicable
   security considerations for the networks involved, the use of an
   overlay enables attacks on encapsulated virtual networks via the
   underlying network.  Examples of such attacks include traffic
   injection into a virtual network via injection of encapsulated
   traffic into the underlying network and modifying underlying network
   traffic to forward traffic among virtual networks that should have no
   connectivity.  The control protocol should provide functionality to
   help counter some of these attacks, e.g., distribution of NVE access
   control lists for each virtual network to enable packets from non-
   participating NVEs to be discarded, but the primary security measures
   for the underlying network need to be applied to the underlying
   network.  For example, if the underlying network includes
   connectivity across the public Internet, use of secure gateways
   (e.g., based on IPsec [RFC4301]) may be appropriate.

   The inner to outer address mappings used for forwarding data towards
   a remote NVE could also be used to filter incoming traffic to ensure

Kreeger, et al.         Expires January 18, 2013               [Page 10]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   the inner address sourced packet came from the correct NVE source
   address, allowing access control to discard traffic that does not
   originate from the correct NVE.  This destination filtering
   functionality should be optional to use.

6.  Acknowledgements

   Thanks to the following people for reviewing and providing feedback:
   Fabio Maino, Victor Moreno, Ajit Sanzgiri, Chris Wright.

7.  Informative References

              Lasserre, M., Balus, F., Morin, T., Bitar, N., and Y.
              Rekhter, "Framework for DC Network Virtualization",
              draft-lasserre-nvo3-framework-03 (work in progress),
              July 2012.

              Narten, T., Sridhavan, M., Dutt, D., Black, D., and L.
              Kreeger, "Problem Statement: Overlays for Network
              draft-narten-nvo3-overlay-problem-statement-03 (work in
              progress), July 2012.

   [RFC4301]  Kent, S. and K. Seo, "Security Architecture for the
              Internet Protocol", RFC 4301, December 2005.

Authors' Addresses

   Lawrence Kreeger


   Dinesh Dutt
   Cumulus Networks


Kreeger, et al.         Expires January 18, 2013               [Page 11]

Internet-Draft   Network Virtualization Overlay CP Reqs        July 2012

   Thomas Narten


   David Black


   Murari Sridharan


Kreeger, et al.         Expires January 18, 2013               [Page 12]