Network Working Group                                         L. Dunbar
Internet Draft                                                 A. Malis                                                Futurewei
Intended status: Informational                                Futurewei                                 A. Malis
Expires: December 19, 2019 May 1, 2020                                        Independent
                                                           C. Jacquenet
                                                                 Orange
                                                          June 19,
                                                       November 1, 2019

           Gap Analysis of Dynamic Networks to Hybrid Cloud DCs
                draft-ietf-rtgwg-net2cloud-gap-analysis-02
                draft-ietf-rtgwg-net2cloud-gap-analysis-03

Abstract

   This document analyzes the technological gaps when using SD-WAN SDWAN to
   dynamically interconnect workloads and applications hosted in
            rd        various 3  party cloud data centers.

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), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-
   Drafts.

   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."

   The list of current Internet-Drafts can be accessed at
   http://www.ietf.org/ietf/1id-abstracts.txt

   The list of Internet-Draft Shadow Directories can be accessed at
   http://www.ietf.org/shadow.html

   This Internet-Draft will expire on December 19, 2019. May 1, 2009.

Copyright Notice

   Copyright (c) 2019 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
   (http://trustee.ietf.org/license-info) 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...................................................2
   2. Conventions used in this document..............................3
   3. Gap Analysis of C-PEs WAN Port Registration....................4 Management......................4
   4. Aggregating VPN paths and Internet paths.......................5 paths.......................6
      4.1. Key Control Plane Components of SD-WAN....................6 SDWAN Overlay.............7
      4.2. Using BGP Tunnel-Encap....................................7 UPDATE Messages.................................8
      4.3. SECURE-L3VPN/EVPN.........................................9
      4.4. Preventing attacks from Internet-facing ports............10
   5. CPEs C-PEs not directly connected to VPN PEs........................10 PEs.......................11
      5.1. Floating PEs to connect to Remote CPEs...................13
      5.2. NAT Traversal............................................13
      5.3. Complexity of using BGP between PEs and remote CPEs via
      Internet......................................................13
      5.4. Designated Forwarder to the remote edges.................14
      5.5. Traffic Path Management..................................15
   6. Manageability Considerations..................................15
   7. Security Considerations.......................................15
   8. IANA Considerations...........................................16
   9. References....................................................16
      9.1. Normative References.....................................16
      9.2. Informative References...................................16
   10. Acknowledgments..............................................17

1. Introduction

   [Net2Cloud-Problem] describes the problems that enterprises face
   today in transitioning their IT infrastructure to support digital
   economy, such as connecting enterprises' branch offices to dynamic
   workloads in different Cloud DCs.

   This document analyzes the technological gaps to interconnect
   dynamic workloads & apps hosted in cloud data centers that the
   enterprise's VPN service provider may not own/operate or may be
   unable to provide the enterprise with the required connectivity to
   access such locations. When VPN service providers have insufficient
   bandwidth to reach a location, SD-WAN SDWAN techniques can be used to
   aggregate bandwidth of multiple networks, such as MPLS VPNs or the
   Public Internet to achieve better performance. This document
   primarily focuses on the technological gaps raised by using SD-WAN SDWAN
   techniques to connect enterprise premises to cloud data centers
   operated by third parties.

   For the sake of readability, a SD-WAN SDWAN edge, a SD-WAN SDWAN endpoint, C-PE,
   or CPE are used interchangeably throughout this document. However,
   each term has some minor emphasis, especially when used in other
   related documents:

     . SD-WAN SDWAN Edge: could include multiple devices (virtual or
        physical);
     . SD-WAN SDWAN endpoint: to refer to a WAN port of SD-WAN SDWAN devices or a
        single SD-WAN SDWAN device;
     . C-PE: more for provider owned SD-WAN SDWAN edge, e.g. for SECURE-
        EVPN's PE based VPN, when PE is the edge node of SD-WAN; SDWAN;
     . CPE: more for enterprise owned SD-WAN SDWAN edge.

2. Conventions used in this document

   Cloud DC:   Third party Data Centers that usually host applications
               and workload owned by different organizations or
               tenants.

   Controller: Used interchangeably with SD-WAN SDWAN controller to manage
               SD-WAN
               SDWAN overlay path creation/deletion and monitor the
               path conditions between sites.

   CPE-Based VPN: Virtual Private Network designed and deployed from
               CPEs. This is to differentiate from most commonly used
               PE-based VPNs a la RFC 4364.

   OnPrem:     On Premises data centers and branch offices

   SD-WAN:

   SDWAN:      Software Defined Wide Area Network, "SD-WAN" "SDWAN" refers to
               the solutions of pooling WAN bandwidth from multiple
               underlay networks to get better WAN bandwidth
               management, visibility & control. When the underlay is a
               private network, traffic may be forwarded without any
               additional encryption; when the underlay networks are
               public, such as the Internet, some traffic needs to be
               encrypted when passing through (depending on user-
               provided policies).

3. Gap Analysis of C-PEs WAN Port Registration

   SD-WAN technology has emerged as means to dynamically and securely
   interconnect Management

   One of the OnPrem branches with key characteristics of the networks that interconnect
   workloads instantiated in Hybrid Cloud DCs is that do not those networks' edges can have
   WAN ports facing networks provided by different ISPs, some can be
   untrusted public internet, some can be MPLS VPN, some can be Cloud
   internal networks, some can be others.

   If an edge node only has one single WAN port facing untrusted
   network, then all sensitive data to/from this edge have direct connectivity to BGP/MPLS VPN PEs be
   encrypted, usually by IPsec tunnels which can be terminated at the
   single WAN port address or at the edge node's internal address if it
   is routable in the wide area network.

   If an edge node has multiple WAN ports with some facing private VPN
   and some facing public untrusted network, sensitive data can be
   forwarded via ports facing VPN natively without encryption and via
   ports facing public network with encryption. To achieve this
   flexibility, it is necessary to have very limited bandwidth. the IPsec tunnels terminated at
   the WAN ports facing the public networks.

   In order to establish pair-wise secure encrypted connection among
   those WAN ports, it is necessary for peers to be informed of the WAN
   port properties.

   Some SD-WAN of those overlay networks (a.k.a. SDWAN in the context of this
   document) use the modified NHRP protocol [RFC2332] to register WAN
   ports of SD-WAN the edges with a their "Controller" (or NHRP server), which
   then has the ability to map a private VPN address to a public IP address of the
   destination node. node/port. DSVPN [DSVPN] or DMVPN [DMVPN] are used to
   establish tunnels between WAN ports of SD-WAN SDWAN edge nodes.

   NHRP was originally intended for ATM address resolution, and as a
   result, it misses many attributes that are necessary for dynamic
   endpoint C-PE registration to the controller, such as:

   - Interworking with the MPLS VPN control plane. A SD-WAN An overlay (SDWAN)
     edge can have some ports facing the MPLS VPN network over which
     packets can be forwarded without any encryption and some ports
     facing the public Internet over which sensitive traffic needs to
     be encrypted before being sent.
   - Scalability: NHRP/DSVPN/DMVPN works fine with small numbers of
     edge nodes. When a network has more than 100 nodes, these
     protocols do not scale well.
   - NHRP does not have the IPsec attributes, which are needed for
     peers to build Security Associations over the public internet.
   - NHRP messages do not have any field to encode the C-PE supported
     encapsulation types, such as IPsec-GRE or IPsec-VxLAN.
   - NHRP messages do not have any field to encode C-PE Location
     identifiers, such as Site Identifier, System ID, and/or Port ID.
   - NHRP messages do not have any field to describe the gateway(s) to
     which the C-PE is attached. When a C-PE is instantiated in a Cloud
     DC, it is desirable for C-PE's owner to be informed of how/where
     the C-PE is attached.
   - NHRP messages do not have any field to describe C-PE's NAT
     properties if the C-PE is using private addresses, such as the NAT
     type, Private address, Public address, Private port, Public port,
     etc.

   [BGP-SDWAN-PORT] describes how SD-WAN SDWAN edge nodes use BGP to register
   their WAN ports properties to the SD-WAN SDWAN controller, which then
   propagates the information to other SD-WAN SDWAN edge nodes that are
   authenticated and authorized to communicate with them.

4. Aggregating VPN paths and Internet paths

   Most likely, enterprises (especially the largest ones) already have
   their CPEs C-PEs interconnected by providers' providers VPNs, based upon VPN
   techniques such as EVPN, L2VPN,
   or L3VPN. The VPN L3VPN, which can be PE-based or CPE-based. The commonly used PE-based PE-
   based VPNs have CPE C-PE directly attached to PEs, therefore the
   communication between CPEs C-PEs and PEs is considered as secure. MP-BGP
   is used to learn & distribute routes among CPEs, C-PEs, even though
   sometimes routes among CPEs C-PEs are statically configured on the CPEs.

   To aggregate paths over the Internet and paths over the VPN, the C-
   PEs need to have some WAN ports connected C-PEs.

   For enterprises already interconnected by VPNs, it is desirable to
   aggregate the PEs of the VPNs bandwidth among VPN paths and
   other WAN Internet paths by C-PEs
   adding additional ports connected facing public internet. Under this scenario
   which is referred to the Internet. It as SDWAN Overlay throughout this document, it
   is necessary for the
   CPEs C-PEs to use a protocol so that they can to register the their WAN
   port properties with their SD-WAN Controller(s): this SDWAN Controller(s). The information
   conditions is
   needed for the establishment and the maintenance of Port-based IPsec
   SA associations among relevant C-PEs.

   When using NHRP for registration purposes, C-PEs need to run two
   separate control planes: EVPN&BGP for CPE-based VPNs, and NHRP &
   DSVPN/DMVPN for ports connected to the Internet. Two separate
   control planes not only add complexity to C-PEs, but also increase
   operational cost.

                                       +---+
                        +--------------|RR |----------+
                       /  Untrusted    +-+-+           \
                      /                                 \
                     /                                   \
     +----+  +---------+  packets encrypted over     +------+  +----+
     | TN3|--|         A1-----+ Untrusted    +------ B1     |--| TN1|
     +----+  | C-PE    A2-\                          | C-PE |  +----+
     +----+  |  A      A3--+--+              +---+---B2  B  |  +----+
     | TN2|--|         |   |PE+--------------+PE |---B3     |--| TN3|
     +----+  +---------+   +--+   trusted    +---+   +------+  +----+
                              |      WAN     |
     +----+  +---------+   +--+   packets    +---+   +------+  +----+
     | TN1|--|         C1--|PE| go natively  |PE |-- D1     |--| TN1|
     +----+  | C-PE    C2--+--+ without encry+---+   | C-PE |  +----+
             |  C      |      +--------------+       |  D   |
             |         |                             |      |
     +----+  |         C3--|  without encrypt over   |      |  +----+
     | TN2|--|         C4--+---- Untrusted  --+------D2     |--| TN2|
     +----+  +---------+                             +------+  +----+
       Figure 1: CPEs interconnected by VPN paths and Internet Paths

 4.1. Key Control Plane Components of SD-WAN SDWAN Overlay

   As described in [BGP-SDWAN-Usage], the SD-WAN SDWAN Overlay Control Plane
   has three distinct properties:

     - SD-WAN SDWAN node's WAN Port Property registration to the SD-WAN SDWAN
        Controller.
          o To inform the SD-WAN SDWAN controller and authorized peers of the
             WAN port properties of the C-PE [SDWAN-Port]. When the WAN
             ports are assigned private addresses, this step can
             register the type of NAT that translates private addresses
             into public ones.

     - Controller facilitated IPsec SA management and NAT information
        distribution
          o It is for SD-WAN SDWAN controller to facilitate or manage the
             IPsec configuration and peer authentication for all IPsec
             tunnels terminated at the SD-WAN SDWAN nodes.

     - Establishing and Managing the topology and reachability for
        services attached to the client ports of SD-WAN SDWAN nodes.

          o This is for the overlay layer's route distribution, so
             that a C-PE can populate its overlay routing table with
             entries that identify the next hop for reaching a specific
             route/service attached to remote nodes. [SECURE-EVPN]
             describes EVPN and other options.

 4.2. Using BGP Tunnel-Encap

   RFC5512 and UPDATE Messages

    [Tunnel-Encap] describe methods to construct the BGP UPDATE
   messages Tunnel Path Attribute that
   advertise endpoints' tunnel encapsulation capability
   and for the
   respective attached client routes, so that routes encoded in the peers that
   receive MP-NLRI Path
   Attribute. The receivers of the BGP UPDATE can establish appropriate tunnels with the
   endpoints for the aforementioined routes. RFC5512 uses the Endpoint
   Address subTLV, whereas [Tunnel-Encap] uses Remote Endpoint Address
   subTLV to indicates address use any of the tunnel endpoint which can be an
   IPv4 or an IPv6 address. There are Tunnel Encapsulation attribute
   subTLVs to indicate the
   supported encapsulation types, such as
   L2TPv3, GRE, VxLAN, IP-in-IP, etc.

   [Tunnel-Encap] removed SAFI =7 (which was specified by RFC5512) encapsulations encoded in the Tunnel Path Attribute for
   distributing encapsulation tunnel information. [Tunnel-Encap]
   requires that tunnels need to be associated with routes.

   There is also
   the Color sub-TLV to describe customer-specified
   information about routes encoded in the tunnels (which can be creatively used for SD-
   WAN). MP-NLRI Path Attribute.

   Here are some of the gaps using [Tunnel-Encap] to control SD-WAN
   Tunnels: distribute SDWAN
   Edge WAN port properties:

   - [Tunnel-Encap] doesn't yet have the functionality that would help the
     C-PE encoding to register its WAN Port properties.
   - A SD-WAN tunnel, e.g. IPsec-based, requires a negotiation between describe the tunnel's end points NAT
     information for supported encryption algorithms and
     tunnel types before it can WAN ports that have private addresses. The NAT
     information needs to be properly established, whereas
     [Tunnel-Encap]  only allow propagated to the announcement of one endpoint's
     supported encapsulation capabilities for specific attached routes
     and no negotiation between tunnel end points is needed. The
     establishment of a SD-WAN tunnel can fail, e.g., in case trusted peers via
     Controllers, such as the two
     endpoints support different encryption algorithms. That virtual C-PEs instantiated in public
     Cloud DCs.
   - It is why a
     SD-WAN tunnel needs not easy using the current mechanism in [Tunnel-Encap] to be established and maintained
     exchange IPsec SA specific parameters independently from
     advertising client routes attached to the edge node.
   - attached clients' routes, even after adding a new
     IPsec tunnel type.
     [Tunnel-Encap] requires all tunnels updates are associated with
     routes. There can be many client routes associated with the SD-WAN SDWAN
     IPsec tunnel between two C-PEs' WAN ports; the corresponding
     destination prefixes (as announced by the aforementioned routes)
     may also be reached through the VPN underlay without any
     encryption. A more realistic approach to separate SD-WAN tunnel
     management from client routes association with the SD-WAN tunnels.
   - When SD-WAN tunnel and clients routes are separate, the SD-WAN
     Tunnel

     The establishment may not have routes associated.
     There is a suggestion on using a "Fake Route" for a SD-WAN node to
     use [Tunnel-Encap] to advertise its SD-WAN of an IPsec tunnel end-points
     properties. However, using "Fake Route" can raise some design
     complexity for large SD-WAN networks with many tunnels. For
     example, for a SD-WAN network with hundreds of nodes, with each
     node having many ports & many endpoints to establish SD-WAN
     tunnels with their corresponding peers, the node would need as
     many "fake addresses". For large SD-WAN networks (such fail, such as those
     comprised of more than 10000 nodes), each node might need 10's
     thousands of "fake addresses", which is very difficult to manage
     and requires lots of configuration tasks due to get the nodes properly
     set up.
   - [Tunnel-Encap] does not have any field to carry detailed
     information about
     two endpoints supporting different encryption algorithms or other
     reasons. There can be multiple negotiations messages for the remote C-PE, such as Site-ID, System-ID,
     Port-ID
   - [Tunnel-Encap] Does not have any field to carry IPsec attributes
     for
     SA parameters between two end points. That is why IPsec SA
     association establishment between end points is independent from
     the SD-WAN edge nodes policies on mapping routes to specific IPSec SA.

     If C-PEs need to establish WAN Port based IPsec Security Associations
     with others. It does not have any proper way for two peer CPEs SA, the
     information encoded in Tunnel Path Attribute should only apply to
     negotiate IPsec keys either, based on
     the configuration sent by WAN ports and should be independent of the Controller.
   - [Tunnel-Encap] does not have any field to indicate clients' routes.

     In addition, the UDP NAT
     private address <-> public address mapping SDWAN Tunnel (IPsec SA) may need to be
     established before clients' routes are attached.

   - C-PEs tend to communicate with a subset of the other C-PEs, not
     all the C-PEs need to be connected through a mesh topology.
     Without any BGP extension, many nodes can get dumped with too much
     Therefore, the distribution of the SDWAN Overlay Edge WAN ports
     information coming from other nodes that they never need be be scoped to
     communicate with. the authorized peers.

4.3. SECURE-L3VPN/EVPN

    [SECURE-L3VPN] describes how to extend the BGP/MPLS VPN [RFC4364]
   capabilities to allow some PEs to connect to other PEs via public
   networks. [SECURE-L3VPN] introduces the concept of Red Interface &
   Black Interface used by PEs, where the RED interfaces are used to
   forward traffic into the VPN, and the Black Interfaces are used
   between WAN ports through which only IPsec-protected packets are
   forwarded to the Internet or to other backbone network thereby
   eliminating the need for MPLS transport in the backbone.

   [SECURE-L3VPN] assumes PEs using MPLS over IPsec when sending
   traffic through the Black Interfaces.

   [SECURE-EVPN] describes a solution where point-to-multipoint BGP
   signaling is used in the control plane for the SDWAN Scenario #1. #1
   described in [BGP-SDWAN-Usage]. It relies upon a BGP cluster design
   to facilitate the key and policy exchange among PE devices to create
   private pair-wise IPsec Security Associations without IKEv2 point-to-point point-
   to-point signaling or any other direct peer-to-peer session
   establishment messages.

   Both [SECURE-L3VPN] and [SECURE-EVPN] are useful, however, they both
   miss the aspects of aggregating VPN and Internet underlays. In
   summary:

   - These Both documents do assume a client traffic is either forwarded all
     encrypted through an IPsec tunnel, or not address the scenario of C-PE having some encrypted at all through
     a different tunnel regardless which WAN ports facing VPN the traffic egress
     the PEs towards WAN. In SDWAN, one client traffic can be forwarded
     encrypted at one time through a WAN port towards untrusted network
     and other ports facing the Internet. be forwarded unencrypted at different time through a WAN port
     to MPLS VPN.

   - The [SECURE-L3VPN] assumes that a CPE "registers" with the RR.
     However, it does not say how. It assumes that the remote CPEs are
     pre-configured with the IPsec SA manually. In SD-WAN, SDWAN, Zero Touch
     Provisioning is expected. Manual configuration is not an option,
     as it contradicts
     especially for the objectives of SD-WAN to automate
     configuration tasks.
   - For RR communication with C-PEs, this draft only mentions IPsec.
     Missing TLS/DTLS. edge devices that are deployed in faraway
     places.

   - The draft [SECURE-L3VPN] assumes that C-PEs and RR are connected with via an
     IPsec tunnel. With zero touch provisioning, we need an Missing TLS/DTLS. The following assumption by
     [SECURE-L3VPN] becomes invalid for SDWAN environment where
     automatic way to
     synchronize the synchronization of IPsec SAs SA between C-PEs and RR. The draft assumes: RR is
     needed:
          A CPE must also be provisioned with whatever additional
          information is needed in order to set up an IPsec SA with
          each of the red RRs

   - IPsec requires periodic refreshment of the keys. The draft does
     not provide any information about how to synchronize the
     refreshment among multiple nodes.
   - IPsec usually sends configuration parameters to two endpoints only
     and lets these endpoints negotiate the key. Let us assume The [SECURE-L3VPN]
     assumes that the RR is responsible for creating creating/managing the key
     for all endpoints: endpoints. When one endpoint is compromised, all other
     connections will be impacted.

4.4. Preventing attacks from Internet-facing ports

   When C-PEs have Internet-facing ports, additional security risks are
   raised.

   To mitigate security risks, in addition to requiring Anti-DDoS
   features on C-PEs, it is necessary for CPEs C-PEs to support means to
   determine whether traffic sent by remote peers is legitimate to
   prevent spoofing attacks.

5. CPEs C-PEs not directly connected to VPN PEs

   Because of the ephemeral property of the selected Cloud DCs, DCs for
   specific workloads/Apps, an enterprise or its network service
   provider may not have direct physical connections to the Cloud DCs
   that are used optimal for hosting the enterprise's specific
   workloads/Apps. Under those circumstances, SD-
   WAN SDWAN Overlay is a very
   flexible choice to interconnect the enterprise on-
   premises on-premises data
   centers & branch offices to its desired Cloud DCs.

   However, SD-WAN SDWAN paths established over the public Internet can have
   unpredictable performance, especially over long distances and across
   operators' domains. Therefore, it is highly desirable to steer as
   much as possible the portion of SD-WAN SDWAN paths over service provider
   VPN (e.g., the enterprise's
   existing VPN) VPN that have has guaranteed SLA to minimize the distance or the
   number of segments over the public Internet.

   MEF Cloud Service Architecture [MEF-Cloud] also describes a use case
   of network operators that uses SD-WAN using SDWAN over LTE or the public Internet for
   last mile access where the VPN service providers cannot necessarily
   provide the required physical infrastructure.

   Under those scenarios, one or two of the SD-WAN SDWAN endpoints may not be
   directly attached to the PEs of a VPN Domain.

   When using SD-WAN SDWAN to connect the enterprise's existing sites with to the
   workloads hosted in Cloud DCs, DCs w, the corresponding CPEs C-PEs have to be
   upgraded to support SD-WAN. SDWAN.  If the workloads hosted in Cloud DCs
   need to be connected to many sites, the upgrade process can be very
   expensive.

   [Net2Cloud-Problem] describes a hybrid network approach that
   integrates SD-WAN SDWAN with traditional MPLS-based VPNs, to extend the
   existing MPLS-based VPNs to the Cloud DC Workloads over the access
   paths that are not under the VPN provider's control. To make it work
   properly, a small number of the PEs of the MPLS VPN can be
   designated to connect to the remote workloads via SD-WAN SDWAN secure IPsec
   tunnels.  Those designated PEs are shown as fPE (floating PE or
   smart PE) in Figure 3. Once the secure IPsec tunnels are
   established, the workloads hosted in Cloud DCs can be reached by the
   enterprise's VPN without upgrading all of the enterprise's existing
   CPEs. The only CPE that needs to support SD-WAN SDWAN would be a
   virtualized CPE instantiated within the cloud DC.

   +--------+                                             +--------+
   | Host-a +--+                                     +----| Host-b |
   |        |  |                                    (')   |        |
   +--------+  |           +-----------+           (   )  +--------+
               |  +-+--+  ++-+        ++-+  +--+-+  (_)
               |  | CPE|--|PE|        |PE+--+ CPE|   |
               +--|    |  |  |        |  |  |    |---+
                  +-+--+  ++-+        ++-+  +----+
                   /       |           |
                  /        |  MPLS   +-+---+    +--+-++--------+
          +------+-+       | Network |fPE-1|    |CPE || Host   |
          | Host   |       |         |     |- --|    ||   d    |
          |   c    |       +-----+   +-+---+    +--+-++--------+
          +--------+       |fPE-2|-----+
                           +---+-+    (|)
                              (|)     (|) SD-WAN SDWAN
                              (|)     (|) over any access
                              +=\======+=========+
                             //   \    | Cloud DC \\
                            //      \ ++-----+       \\
                                      +Remote|
                                      |  CPE |
                                      +-+----+
                            ----+-------+-------+-----
                                |               |
                            +---+----+      +---+----+
                            | Remote |      | Remote |
                            | App-1  |      | App-2  |
                            +--------+      +--------+

                    Figure 3: VPN Extension to Cloud DC

   In Figure 3, the optimal Cloud DC to host the workloads (as a
   function of the proximity, capacity, pricing, or other criteria
   chosen by the enterprises) does not have a direct connection to the
   PEs of the MPLS VPN that interconnects the enterprise's existing
   sites.

5.1. Floating PEs to connect to Remote CPEs

   To extend MPLS VPNs to remote CPEs, it is necessary to establish
   secure tunnels (such as IPsec tunnels) between the Floating PEs and
   the remote CPEs.

   Even though a set of PEs can be manually selected to act as the
   floating PEs for a specific cloud data center, there are no standard
   protocols for those PEs to interact with the remote CPEs (most
   likely virtualized) instantiated in the third party cloud data
   centers (such as exchanging performance or route information).

   When there is more than one fPE available for use (as there should
   be for resiliency purposes or the ability to support multiple cloud
   DCs geographically scattered), it is not straightforward to
   designate an egress fPE to remote CPEs based on applications.  There
   is too much applications' traffic traversing PEs, and it is not
   feasible for PEs to recognize applications from the payload of
   packets.

5.2. NAT Traversal

   Cloud DCs that only assign private IPv4 addresses to the
   instantiated workloads assume that traffic to/from the workload
   usually needs to traverse NATs.
   A SD-WAN SDWAN edge node can solicit a STUN (Session Traversal of UDP
   Through Network Address Translation RFC 3489) Server to get the NAT
   property, the public IP address and the Public Port number so that
   such information can be communicated to the relevant peers.

5.3. Complexity of using BGP between PEs and remote CPEs via Internet

   Even though an EBGP (external BGP) Multi-hop design can be used to
   connect peers that are not directly connected to each other, there
   are still some complications in extending BGP from MPLS VPN PEs to
   remote CPEs via any access path (e.g., Internet).

   The path between the remote CPEs and VPN PEs that maintain VPN
   routes may very well traverse untrusted nodes.

   EBGP Multi-hop design requires static configuration on both peers.
   To use EBGP between a PE and remote CPEs, the PE has to be manually
   configured with the "next-hop" set to the IP address of the CPEs.
   When remote CPEs, especially remote virtualized CPEs are dynamically
   instantiated or removed, the configuration of Multi-Hop EBGP on the
   PE has to be changed accordingly.

     Egress peering engineering (EPE) is not sufficient. Running BGP on
     virtualized CPEs in Cloud DCs requires GRE tunnels to be
     established first, which requires the remote CPEs to support
     address and key management capabilities. RFC 7024 (Virtual Hub &
     Spoke) and Hierarchical VPN do not support the required
     properties.

     Also, there is a need for a mechanism to automatically trigger
     configuration changes on PEs when remote CPEs' are instantiated or
     moved (leading to an IP address change) or deleted.

     EBGP Multi-hop design does not include a security mechanism by
     default. The PE and remote CPEs need secure communication channels
     when connecting via the public Internet.

   Remote CPEs, if instantiated in Cloud DCs, might have to traverse
   NATs to reach PEs. It is not clear how BGP can be used between
   devices located beyond the NAT and the devices located behind the
   NAT. It is not clear how to configure the Next Hop on the PEs to
   reach private IPv4 addresses.

5.4. Designated Forwarder to the remote edges

   Among the multiple floating PEs that are reachable from a remote
   CPE, multicast traffic sent by the remote CPE towards the MPLS VPN
   can be forwarded back to the remote CPE due to the PE receiving the
   multicast packets forwarding the multicast/broadcast frame to other
   PEs that in turn send to all attached CPEs. This process may cause
   traffic loops.

   Therefore, it is necessary to designate one floating PE as the CPE's
   Designated Forwarder, similar to TRILL's Appointed Forwarders
   [RFC6325].

   MPLS VPNs do not have features like TRILL's Appointed Forwarders.

5.5. Traffic Path Management

   When there are multiple floating PEs that have established IPsec
   tunnels with the remote CPE, the remote CPE can forward outbound
   traffic to the Designated Forwarder PE, which in turn forwards
   traffic to egress PEs and then to the final destinations. However,
   it is not straightforward for the egress PE to send back the return
   traffic to the Designated Forwarder PE.

   Example of Return Path management using Figure 3 above.

   - fPE-1 is DF for communication between App-1 <-> Host-a due to
   latency, pricing or other criteria.
   - fPE-2 is DF for communication between App-1 <-> Host-b.

6. Manageability Considerations

     Zero touch provisioning of SD-WAN SDWAN edge nodes should be a major
     feature of SD-WAN SDWAN deployments. It is necessary for a newly powered
     up SD-WAN SDWAN edge node to establish a secure connection (by means of
     TLS, DTLS, etc.) with its controller.

7. Security Considerations

     The intention of this draft is to identify the gaps in current and
     proposed SD-WAN SDWAN approaches that can address requirements identified
     in [Net2Cloud-problem].

     Several of these approaches have gaps in meeting enterprise
     security requirements when tunneling their traffic over the
     Internet, since this is the purpose of SD-WAN. SDWAN. See the individual
     sections above for further discussion of these security gaps.

8. IANA Considerations

   This document requires no IANA actions. RFC Editor: Please remove
   this section before publication.

9. References

9.1. Normative References

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

9.2. Informative References

   [RFC8192] S. Hares, et al, "Interface to Network Security Functions
             (I2NSF) Problem Statement and Use Cases", July 2017

   [RFC5521] P. Mohapatra, E. Rosen, "The BGP Encapsulation Subsequent
             Address Family Identifier (SAFI) and the BGP Tunnel
             Encapsulation Attribute", April 2009.

   [BGP-SDWAN-PORT]L. Dunbar, et al, "Subsequent Address Family
             Indicator for SDWAN Ports", draft-dunbar-idr-sdwan-port-
             safi-00, Work-in-progress, March 2019.

   [BGP-SDWAN-Usage] L. Dunbar, et al, "Framework of Using BGP for
             SDWAN Overlay Networks", draft-dunbar-idr-sdwan-framework-
             00, work-in-progress, Feb 2019.

   [Tunnel-Encap]E. Rosen, et al, "The BGP Tunnel Encapsulation
             Attribute", draft-ietf-idr-tunnel-encaps-10, July 2018.

   [SECURE-EVPN A. Sajassi, et al, draft-sajassi-bess-secure-evpn-01,
             work in progress, March 2019.

   [SECURE-L3VPN] E. Rosen, "Provide Secure Layer L3VPNs over Public
             Infrastructure", draft-rosen-bess-secure-l3vpn-00, work-
             in-progress, July 2018

   [DMVPN] Dynamic Multi-point VPN:
             https://www.cisco.com/c/en/us/products/security/dynamic-
             multipoint-vpn-dmvpn/index.html

   [DSVPN] Dynamic Smart VPN:
             http://forum.huawei.com/enterprise/en/thread-390771-1-
             1.html

   [ITU-T-X1036] ITU-T Recommendation X.1036, "Framework for creation,
             storage, distribution and enforcement of policies for
             network security", Nov 2007.

    [Net2Cloud-Problem] L. Dunbar and A. Malis, "Seamless Interconnect
             Underlay to Cloud Overlay Problem Statement", draft-dm-
             net2cloud-problem-statement-02, June 2018

10. Acknowledgments

   Acknowledgements to xxx John Drake for his review and contributions.
   Many thanks to John Scudder for stimulating the clarification
   discussion on the Tunnel-Encap draft so that our gap analysis can be
   more accurate.

   This document was prepared using 2-Word-v2.0.template.dot.

Authors' Addresses

   Linda Dunbar
   Futurewei
   Email: ldunbar@futurewei.com

   Andrew G. Malis
   Futurewei
   Independent
   Email: agmalis@gmail.com

   Christian Jacquenet
   Orange
   Rennes, 35000
   France
   Email: Christian.jacquenet@orange.com