Skip to main content

Dynamic Networks to Hybrid Cloud DCs: Problems and Mitigation Practices
draft-ietf-rtgwg-net2cloud-problem-statement-37

Document Type Active Internet-Draft (rtgwg WG)
Authors Linda Dunbar , Andrew G. Malis , Christian Jacquenet , Mehmet Toy , Kausik Majumdar
Last updated 2024-03-17
Replaces draft-dm-net2cloud-problem-statement
RFC stream Internet Engineering Task Force (IETF)
Intended RFC status (None)
Formats
Reviews
Additional resources Mailing list discussion
Stream WG state In WG Last Call
Awaiting External Review/Resolution of Issues Raised, Doc Shepherd Follow-up Underway
Document shepherd Joel M. Halpern
Shepherd write-up Show Last changed 2024-02-20
IESG IESG state I-D Exists
Consensus boilerplate Unknown
Telechat date (None)
Responsible AD (None)
Send notices to jmh@joelhalpern.com
draft-ietf-rtgwg-net2cloud-problem-statement-37
RTG Working Group                                             L. Dunbar
Internet Draft                                                Futurewei
Intended status: Informational                                 A. Malis
Expires: September 18, 2024                            Malis Consulting
                                                           C. Jacquenet
                                                                 Orange
                                                                 M. Toy
                                                                Verizon
                                                            K. Majumdar
                                                              Microsoft
                                                          March 18, 2024

       Dynamic Networks to Hybrid Cloud DCs: Problems and Mitigation
                                 Practices
              draft-ietf-rtgwg-net2cloud-problem-statement-37

Abstract

   This document describes a set of network-related problems
   enterprises face at the time of writing this document (2023) when
   interconnecting their branch offices with dynamic workloads in
   third-party data centers (DCs) (a.k.a. Cloud DCs). These problems
   are mainly from enterprises with conventional VPN services that want
   to leverage those networks (instead of altogether abandoning them).
   This document also describes various mitigation practices and
   actions to soften the issues induced by these problems.

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

Dunbar, et al.                                                 [Page 1]
Internet-Draft     Net2Cloud Problems & Mitigations

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

   This Internet-Draft will expire on September 4, 2024.

Copyright Notice

   Copyright (c) 2024 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...................................................3
   2. Definition of Terms............................................3
   3. Issues and Mitigation Methods of Connecting to Cloud DCs.......4
      3.1. Increased BGP Peering Errors and Mitigation Methods.......4
      3.2. Site Failures and Methods to Minimize Impacts.............6
      3.3. Limitations of DNS-based Cloud DC Location Selection......6
      3.4. Network Issues for 5G Edge Clouds and Mitigation Methods..7
      3.5. DNS Practices for Hybrid Workloads........................8
      3.6. NAT Practices for Accessing Cloud Services................9
      3.7. Cloud Discovery Practices.................................9
   4. Dynamic Connecting Enterprise Sites with Cloud DCs............10
      4.1. Sites to Cloud DC........................................10
      4.2. Inter-Cloud Connection...................................12
      4.3. Extending Private VPNs to Hybrid Cloud DCs...............14
   5. Methods to Scale IPsec Tunnels to Cloud DCs...................15
      5.1. Scale IPsec Tunnels Management...........................15
      5.2. CPEs Interconnection Over the Public Internet............16
   6. Requirements for Networks Connecting Cloud Data Centers.......16
   7. Security Considerations.......................................17
   8. IANA Considerations...........................................19
   9. References....................................................19
      9.1. Normative References.....................................19

Dunbar, et al.                                                 [Page 2]
Internet-Draft     Net2Cloud Problems & Mitigations

      9.2. Informative References...................................20
   10. Acknowledgments..............................................21

1. Introduction
   With the advent of widely available Cloud data centers (DCs)
   providing services in various geographic locations and advanced
   tools for monitoring and predicting application behaviors, it is
   tempting for enterprises to instantiate applications and workloads
   in Cloud DCs. Some enterprises prefer specific applications to be
   located close to the end users accessing these services, as the
   proximity can improve end-to-end latency. In addition, applications
   and workloads in Cloud DCs can be shut down or moved along with end
   users in motion thereby modifying the networking connection of
   subsequently relocated applications and workloads.
   Cloud services are generally exposed, on-demand services that claim
   to be scalable, highly available, and have usage-based billing. Most
   Cloud Operators provide Cloud network functions, such as virtual
   Firewall services, virtual private clouds services, and virtual
   Private Branch eXchange (PBX) services, including voice and video
   conferencing systems. A Cloud DC is a shared infrastructure that
   hosts services to many customers.
   This document describes the network-related problems enterprises
   face at the time of writing this document when interconnecting their
   branch offices with dynamic workloads in Cloud DCs and the
   mitigation practices to get around those problems.

2. Definition of Terms

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

   Heterogeneous Cloud: applications and workloads split among Cloud
               DCs owned or managed by different operators.

   Hybrid Clouds: A hybrid cloud is a mixed computing environment where
               applications are run using a combination of computing,
               storage, and services in different public clouds and
               private clouds, including on-premises data centers or

Dunbar, et al.                                                 [Page 3]
Internet-Draft     Net2Cloud Problems & Mitigations

               "edge" locations. <https://cloud.google.com/learn/what-
               is-hybrid-cloud>.

   IXPs:       Internet exchange points (IXes or IXPs) are the common
               grounds of IP networking, allowing participating
               Internet service providers (ISPs) to exchange data
               destined for their respective networks.
               <https://en.wikipedia.org/wiki/Internet_exchange_point>.

   SD-WAN      An overlay connectivity service that optimizes transport
               of IP Packets over one or more Underlay Connectivity
               Services by recognizing applications (Application Flows)
               and determining forwarding behavior by applying Policies
               to them. [MEF-70.1]

   VPC:        A Virtual Private Cloud is a virtual network dedicated
               to one client account. It is logically isolated from
               other virtual networks in a Cloud DC. Each client can
               launch his/her desired resources, such as compute,
               storage, or network functions into his/her VPC. At the
               time of writing this document, most Cloud operators'
               VPCs only support private addresses, some support IPv4
               only, others support IPv4/IPv6 dual stack.

3. Issues and Mitigation Methods of Connecting to Cloud DCs

   This section identifies some high-level problems that the IETF could
   address, especially within the Routing area. Other Cloud DC problems
   (e.g., managing cloud spending) are out of the scope of this
   document.

3.1. Increased BGP Peering Errors and Mitigation Methods

   Where conventional ISPs view BGP peering as a means to improve
   network operations, Public Cloud DCs offer direct BGP peering to
   attract more customers to use their DCs and services. As such, there
   is pressure to peer more widely with more customers, including those
   who may lack the expertise and experience in running complex BGP
   peering relationships. This can contribute to increased BGP peering
   errors such as capability mismatch, unwanted route leaks, missing
   Keepalives, and errors causing BGP session resets. Capability

Dunbar, et al.                                                 [Page 4]
Internet-Draft     Net2Cloud Problems & Mitigations

   mismatch can cause BGP sessions not to be adequately established.
   These issues are more acute for Cloud DCs than they have been, even
   though they may apply to conventional ISPs, just to a lesser degree.
   Here are the recommended mitigation practices:

     - If a Cloud Gateway (GW), a BGP speaker, receives from its BGP
        peer a capability that it does not itself support or recognize,
        it need to ignore that capability, and the BGP session need not
        be terminated per [RFC5492]. When receiving a BGP UPDATE with a
        malformed attribute, the revised BGP error handling procedure
        in [RFC7606] should be followed instead of session resetting.
     - When a Cloud DC doesn't support multi-hop eBGP peering with
        external devices, as many don't, enterprise GWs need to
        establish tunnels (e.g., IPsec) to the Cloud GWs to form an IP
        adjacency.
     - When a Cloud DC eBGP session supports a limited number of
        routes from external entities, the on-premises DCs need to set
        up default routes and filter as many routes as practical
        replacing them with a default in the eBGP advertisement to
        minimize the number of routes to be exchanged with the Cloud DC
        eBGP peers.
     - When a Cloud GW receives inbound routes exceeding the maximum
        routes threshold for a peer, the currently common practice is
        generating out-of-band alerts (e.g., Syslog entries) via the
        management system or terminating the BGP session (with cease
        notification messages [RFC4486] being sent). Although out of
        the scope of this document, more discussion is needed in the
        IETF Inter-Domain Routing (IDR) Working Group for potential in-
        band or autonomous notification directly to the peers when the
        inbound routes exceed the maximum routes threshold.
     - Leveraging YANG models to programmatically synchronize
        configurations between BGP peers (e.g., [SVC-AC]) and to adjust
        the local configuration accordingly (e.g., [NTW-AC] or
        [DATAMODEL-BGP]). This proactive approach reduces the
        likelihood of BGP configuration issues and ensures that both
        BGP peers operate with synchronized and compatible settings.

Dunbar, et al.                                                 [Page 5]
Internet-Draft     Net2Cloud Problems & Mitigations

3.2. Site Failures and Methods to Minimize Impacts

   Failures within a Cloud site, which can be a building, a floor, a
   pod, or a server rack, include capacity degradation or complete out-
   of-service failure. Here are some events that can trigger a site
   failure: a) fiber cut for links connecting to the site or among pods
   within the site; b) cooling failures; c) insufficient backup power
   during a power failure; d) cyber threat attacks; e) too many changes
   outside of the maintenance window; etc. A fiber-cut is not uncommon
   in a Cloud site or between sites.

   As described in [RFC7938], a Cloud DC might not have an IGP to route
   around link/node failures within its domain. When a site failure
   happens, the Cloud DC GW visible to clients is running fine;
   therefore, the site failure is not detectable by the clients using
   Bidirectional Forwarding Detection (BFD)[RFC5880].

   When a site failure occurs, many services can be impacted. When the
   impacted services' IP prefixes in a Cloud DC are not aggregated
   nicely, which is common, one single site failure can trigger a huge
   number of BGP UPDATE messages. There are proposals, such as
   [METADATA-PATH], to enhance BGP advertisements to address this
   problem.

   [RFC7432] specifies a mass withdrawal mechanism for EVPN to signal a
   large number of routes being changed to remote PE nodes as quickly
   as possible.

3.3. Limitations of DNS-based Cloud DC Location Selection

   Many applications have multiple instances running in different Cloud
   DCs. A commonly deployed solution has DNS server(s) responding to a
   Fully Qualified Domain Name (FQDN) inquiry with an IP address of the
   instance in the closest or lowest cost DC. Here are some problems
   associated with DNS-based solutions:
     - Dependent on client behavior
          - A misbehaving client can cache results indefinitely.
          - Clients may fail to access a service even though there are
             servers available in other Cloud DCs because the failing
             IP address is still cached in the DNS resolver and has not
             expired yet.

Dunbar, et al.                                                 [Page 6]
Internet-Draft     Net2Cloud Problems & Mitigations

     - No inherent use of proximity information present in the network
        (routing) layer, resulting in loss of performance.
     - Inflexible traffic control:
        The Local DNS resolver becomes the unit of traffic management.
        This requires DNS to receive periodic updates of the network
        condition, which is difficult.

   One method to mitigate the problems listed above is to use anycast
   [RFC4786] for the services so that network proximity and conditions
   can be automatically considered in optimal path selection.

   [METADATA-PATH] identifies some of the metrics that can be utilized
   for the ingress routers to make path steering selections not only
   based on the routing cost but also the running environment of the
   edge services.

   [RFC8490] and [RFC8765] on stateful DNS can be used to achieve
   better performance in refreshing the cache and handling session idle
   timeouts.

3.4. Network Issues for 5G Edge Clouds and Mitigation Methods

   5G Edge Cloud DCs [3GPP-5G-Edge] may host edge computing
   applications for ultra-low latency services on virtual or physical
   servers. Those edge computing applications have low latency
   connections to the UEs (User Equipment) and might have other
   connections to backend servers or databases in other locations.

   The low latency traffic to/from the UEs is transported through the
   5G Core (gNB (Next Generation Node B))<-> UPFs (User Plane
   Function)) and the 5G Local Data Networks (LDN) to the edge Cloud
   DCs. The LDN's ingress routers connected to the UPFs might be co-
   located with 5G Core functions in the edge Clouds. The 5G Core
   functions include Radio Control Functions, Session Management
   Functions (SMF), Access Mobility Functions (AMF), User Plane
   Functions (UPF), and others.

   Here are some network problems with connecting to the services in
   the 5G Edge Clouds:

       1) The difference in routing distances to server instances in
          different edge Clouds is relatively small. Therefore, the
          instance in the Edge Cloud with the shortest routing distance

Dunbar, et al.                                                 [Page 7]
Internet-Draft     Net2Cloud Problems & Mitigations

          from a 5G UPF might not be the best in providing the overall
          low latency service.
       2) Capacity status at the Edge Cloud might play a more
          significant role in end-to-end performance.
       3) Source (UEs) can ingress from different LDN Ingress routers
          due to mobility.

   [METADATA-PATH] describes a mechanism to get around those problem.
   [METADATA-PATH] extends the BGP UPDATE messages for a Cloud GW to
   propagate the edge service-related metrics from Cloud GW to the
   ingress routers so that the ingress routers can incorporate the
   destination site's capabilities with the routing distance in
   computing the optimal paths.

   The IETF CATS (Computing-Aware Traffic Steering) working group is
   examining general aspects of this space, and may come up with
   protocol recommendations for this information exchange.

3.5. DNS Practices for Hybrid Workloads

   DNS name resolution is essential for on-premises and cloud-based
   resources. For customers with hybrid workloads, which include on-
   premises and cloud-based resources, extra steps are necessary to
   configure DNS to work seamlessly across both environments.

   Cloud operators have their own DNS to resolve resources within their
   Cloud DCs and to well-known public domains. Cloud's DNS can be
   configured to forward queries to customer managed authoritative DNS
   servers hosted on-premises and to respond to DNS queries forwarded
   by on-premises DNS servers.

   For enterprises utilizing Cloud services provided by different Cloud
   operators, it is necessary to establish policies and rules on
   how/where to forward DNS queries. When applications in one Cloud
   need to communicate with applications hosted in another Cloud, DNS
   queries from one Cloud DC could be forwarded to the enterprises' on-
   premises DNS, which in turn can be forwarded to the DNS service in
   another Cloud. Configuration can be complex depending on the
   application communication patterns.

   However, name collisions can still occur even with carefully managed
   policies and configurations. If an organization uses internal names
   like those under a .internal top level domain name, and wants its
   services to be available via or within some other Cloud provider

Dunbar, et al.                                                 [Page 8]
Internet-Draft     Net2Cloud Problems & Mitigations

   that also uses .internal, collisions might occur. Therefore, using a
   global domain name is better even when an organization does not make
   all its namespace globally resolvable. An organization's globally
   unique DNS can include subdomains that cannot be resolved outside
   certain restricted paths, zones that resolve differently based on
   the origin of the query, and zones that resolve the same globally
   for all queries from any source [Split-Horizon-DNS].

   Globally unique names do not equate to globally resolvable names or
   even global names that resolve the same way from every perspective.
   Globally unique names can prevent any possibility of collisions, and
   they make DNSSEC trust manageable. Consider using a registered and
   FQDN from global DNS as the root for enterprise and other internal
   namespaces.

3.6. NAT Practices for Accessing Cloud Services

   Cloud resources, such as VMs (Virtual Machine) or application
   instances, are commonly assigned with private IP addresses. By
   configuration, some private subnets can have NAT functionality to
   reach out to external networks, and some private subnets are
   internal to a Cloud DC only.

   Different Cloud operators support different levels of NAT
   functionality. For example, AWS NAT Gateway does not currently
   support connections towards, or from, VPC Endpoints, VPN, AWS Direct
   Connect, or VPC Peering [AWS-NAT]. AWS Direct Connect/VPN/VPC
   Peering does not currently support any NAT functionality.

   Google's Cloud NAT [Google-NAT] allows Google Cloud VM instances
   without external IP addresses and private Google Kubernetes Engine
   (GKE) clusters to connect to the Internet. Cloud NAT implements
   outbound NAT in conjunction with a default route to allow instances
   to reach the Internet. It does not implement inbound NAT. Hosts
   outside the VPC network can only respond to established connections
   initiated by instances inside the Google Cloud; they cannot initiate
   new connections to Cloud instances via NAT.

   For enterprises with applications running in different Cloud DCs,
   proper configuration of NAT need to be performed in Cloud DCs and
   their on-premises DC.

3.7. Cloud Discovery Practices

   One of the concerns of enterprises using Cloud services is the lack
   of awareness of the locations of their services hosted in the Cloud,

Dunbar, et al.                                                 [Page 9]
Internet-Draft     Net2Cloud Problems & Mitigations

   as Cloud operators can move the service instances from one place to
   another. While the geographic locations are usually exposed to the
   enterprises, such as Availability Zones or Regions, the topological
   location is usually hidden. When applications in Cloud DCs
   communicate with on-premises applications, it may not be clear where
   the Cloud applications are located or to which VPCs they belong.

   Being able to detect Cloud services' location can help on-premises
   gateways (routers) to connect to services in a more optimal site
   when the enterprise's end users or policies change.

   For enterprises that instantiate virtual routers in Cloud DCs,
   metadata can be attached (e.g., GENEVE [RFC8926] header or IPv6
   optional header) to indicate additional properties, including useful
   information about the sites where they are instantiated.

4. Dynamic Connecting Enterprise Sites with Cloud DCs

   For many enterprises with established private VPNs (e.g., private
   circuits, MPLS-based L2VPN[RFC6136]/L3VPN[RFC4364]) interconnecting
   branch offices and on-premises data centers, connecting to Cloud
   services will be a mix of different types of networks. When an
   enterprise's existing VPN service providers do not have direct
   connections to the desired cloud DCs that the enterprise prefers to
   use, the enterprise faces additional infrastructure and operational
   costs to utilize the Cloud services.

   This section describes some mechanisms for enterprises with private
   VPNs to connect to Cloud services dynamically.

4.1. Sites to Cloud DC

   Most Cloud operators offer multiple types of network gateways (GWs)
   through which an enterprise can reach their workloads hosted in the
   Cloud DCs:

     - Internet GW for services hosted in the Cloud DCs to be accessed
        by external requests via Internet routable addresses. E.g., AWS
        Internet GW [AWS-Cloud-WAN].
     - IPsec tunnels terminating GW for establishing IPsec SAs
        [RFC6071] with an enterprise's own gateway, so that the
        communications between those gateways can be secured from the

Dunbar, et al.                                                [Page 10]
Internet-Draft     Net2Cloud Problems & Mitigations

        underlay (which might be the public Internet). E.g., AWS
        Virtual gateway (vGW).
     - Direct connect GW for enterprises to connect with Cloud
        services via private leased lines provided by Network Service
        Providers. E.g., AWS Direct Connect. In addition, an AWS Transit
        Gateway can be used to interconnect multiple VPCs in different
        Availability Zones. AWS Transit Gateway acts as a hub that
        controls how traffic is forwarded among all the connected
        networks which act like spokes.,

   Microsoft Azure's Virtual WAN [Azure-SD-WAN] allows extension of a
   private network to any of the Microsoft Cloud services, including
   Azure and Office365. ExpressRoute is configured using Layer 3
   routing. Customers can opt for redundancy by provisioning dual links
   from their location to two Microsoft Enterprise edge routers (MSEEs)
   located within a third-party ExpressRoute peering location. The BGP
   routing protocol is then setup over WAN links to provide redundancy
   to the cloud. This redundancy is maintained from the peering data
   center into Microsoft's cloud network.

   Google's Cloud Dedicated Interconnect offers similar network
   connectivity options as AWS and Microsoft. One distinct difference,
   however, is that Google's service allows customers access to the
   entire global Cloud network by default. It does this by connecting
   the on-premises network with the Google Cloud using BGP and Google
   Cloud Routers to provide optimal paths to the different regions of
   the global cloud infrastructure.

   Figure 1 below shows an example of a portion of workloads belonging
   to one tenant (e.g., TN-1) that are accessible via a virtual router
   connected by AWS Internet Gateway; some of the same tenant (TN-1)
   services are accessible via AWS vGW, and others are accessible via
   AWS Direct Connect. The workloads belonging to one tenant can
   communicate within a Cloud DC via virtual routers (e.g., vR1, vR2).

   Different types of access require different level of security
   functions. Sometimes it is not visible to end customers which type
   of network access is used for a specific application instance.  To
   get better visibility, separate virtual routers (e.g., vR1 & vR2)
   can be deployed to differentiate traffic to/from different Cloud

Dunbar, et al.                                                [Page 11]
Internet-Draft     Net2Cloud Problems & Mitigations

   GWs. It is important for some enterprises to be able to observe the
   specific behaviors when connected by different connections.

   A CPE (Customer Premises Equipment) can be a customer owned router
   or ports physically connected to an AWS Direct Connect GW.
     +------------------------+
     |    ,---.         ,---. |
     |   (TN-1 )       ( TN-2)|
     |    `-+-'  +---+  `-+-' |
     |      +----|vR1|----+   |
     |           ++--+        |
     |            |         +-+----+
     |            |        /Internet\ For external customers
     |            +-------+ Gateway  +----------------------
     |                     \        / to reach via Internet
     |                      +-+----+
     |                        |
     |    ,---.         ,---. |
     |   (TN-1 )       ( TN-2)|
     |    `-+-'  +---+  `-+-' |
     |      +----|vR2|----+   |
     |           ++--+        |
     |            |         +-+----+
     |            |        / virtual\ For IPsec Tunnel
     |            +-------+ Gateway  +----------------------
     |            |        \        /  termination
     |            |         +-+----+
     |            |           |
     |            |    + - - - - - - - - - - - - - - - --+
     |            |    |    +-+----+          +----+
     |            |        /        \ Direct /      \   |
     |            +----|--+ Gateway  +------+ Fabric|--VPN-- CPE
     |                     \        / Connect\ edge /   |
     |                 |    +-+----+          +----+
     |                        |         IXP              |
     |                 + - - - - - - - - - - - - - - - --+
     +------------------------+
     TN: Tenant Network. One TN can be attached to both vR1 and vR2.
     Figure 1: Examples of Multiple Cloud DC connections.

4.2. Inter-Cloud Connection

   The connectivity options to Cloud DCs described in Section 4.1 are
   for reaching Cloud providers' DCs, but not between cloud DCs. For
   example, when applications in AWS Cloud need to communicate with
   applications in Azure, today's practice requires a third-party

Dunbar, et al.                                                [Page 12]
Internet-Draft     Net2Cloud Problems & Mitigations

   gateway (physical or virtual) to interconnect the AWS's Layer 2
   DirectConnect path with Azure's Layer 3 ExpressRoute.

   Enterprises can also instantiate their virtual routers in different
   Cloud DCs and administer IPsec tunnels among them. In summary, here
   are some approaches, available to interconnect workloads among
   different Cloud DCs:

     a) Utilize Cloud DC provided inter/intra-cloud connectivity
        services (e.g., AWS Transit Gateway) to connect workloads
        instantiated in multiple VPCs. Such services are provided with
        the Cloud gateway to connect to external networks (e.g., AWS
        DirectConnect Gateway).
     b) Hairpin all traffic through the customer gateway, meaning all
        workloads are directly connected to the customer gateway, so
        that communications among workloads within one Cloud DC must
        traverse the customer gateway.
     c) Establish direct tunnels among different VPCs (AWS' Virtual
        Private Clouds) and VNET (Azure's Virtual Networks) via
        client's own virtual routers instantiated within Cloud DCs.
        NHRP (Next Hop Resolution Protocol) [RFC2735] based multi-point
        techniques can be used to establish direct multi-point-to-Point
        or multi-point-to multi-point tunnels among those client's own
        virtual routers.
     d) Utilize a Cloud Aggregator or Cloud Services Broker (CSB) who
        acts as an intermediary among cloud service providers and
        network service providers to offer a combined total package for
        enterprises. The Cloud Aggregator can provide the network
        connections among one enterprise's services instantiated in
        multiple Clouds.

   Approach a) usually does not work if Cloud DCs are owned and managed
   by different Cloud providers.

   Approach b) creates additional transmission delay plus incurring
   costs when exiting Cloud DCs.

   For Approach c), [SDWAN-EDGE-DISCOVERY] describes a mechanism for
   virtual routers to advertise their properties for establishing
   proper IPsec tunnels among them. There could be other approaches
   developed to address the problem.

Dunbar, et al.                                                [Page 13]
Internet-Draft     Net2Cloud Problems & Mitigations

   Approach d) is a method of third-party multi-cloud management
   business model.

4.3. Extending Private VPNs to Hybrid Cloud DCs

   Traditional private VPNs, including private circuits or MPLS-based
   L2/L3 VPNs, have been widely deployed as an effective way to support
   businesses and organizations that require network performance and
   reliability although such services may be considered premium,
   available only at additional cost. Connecting an enterprise's on-
   premesis CPEs to a Cloud DC via a private VPN requires the private
   VPN provider to have a direct path to the Cloud GW. When the user
   base changes, the enterprise might want to migrate its
   workloads/applications to a new cloud DC location closer to the new
   user base. The existing private VPN provider might not have circuits
   at the new location. Deploying PEs routers at new locations takes a
   long time (weeks, if not months).

   When the private VPN network can't reach the desired Cloud DCs,
   IPsec tunnels can dynamically connect the private VPN's PEs with the
   desired Cloud DCs GWs. As the private VPNs provide higher quality of
   services, choosing a PE closest to the Cloud GW for the IPsec tunnel
   is desirable to minimize the IPsec tunnel distance over the public
   Internet.

   In order to support Explicit Congestion Notification (ECN) [RFC3168]
   usage by private VPN traffic, the PEs that establish the IPsec
   tunnels with the Cloud GW need to comply with the ECN behavior
   specified by [RFC6040].

   An enterprise can connect to multiple Cloud DC locations and
   establish different BGP peering with Cloud GW routers at different
   locations. As multiple Cloud DCs are interconnected by the Cloud
   provider's own internal network, its topology and routing policies
   are not transparent or even visible to the enterprise customer's on-
   premises routers. One Cloud GW BGP session might advertise all of
   the prefixes of the enterprise's VPC, regardless of which Cloud DC a
   given prefix resides, which can cause improper optimal path
   selection for on-premises routers. To get around this problem,
   virtual routers in Cloud DCs can be used to attach metadata (e.g.,
   in the GENEVE header or IPv6 optional header) to indicate the Geo-

Dunbar, et al.                                                [Page 14]
Internet-Draft     Net2Cloud Problems & Mitigations

   location of the Cloud DC, the delay measurement, or other relevant
   data.

5. Methods to Scale IPsec Tunnels to Cloud DCs

   As described in Section 4.3, IPsec tunnels can be used to
   dynamically establish connection between private VPN PEs with Cloud
   GWs. Enterprises can also instantiate virtual routers within Cloud
   DCs to connect to their on-premises devices via IPsec tunnels.

   As described in [Int-tunnels], IPsec tunnels can introduce MTU
   problems. This document assumes that endpoints manage the
   appropriate MTU sizes, therefore, not requiring VPN PEs to perform
   fragmentation when encapsulating user payloads in the IPsec packets.

5.1. Scale IPsec Tunnels Management

   IPsec tunnels are a very convenient solution for an enterprise with
   a small number of locations to reach a Cloud DC. However, for a
   medium-to-large enterprise with multiple sites and data centers to
   fully connect to multiple cloud DCs, there are N*C*2 bi-directional
   IPsec SAs (tunnels) between Cloud DC gateways and all those sites,
   with N being the number of enterprise sites and C being the number
   of Cloud sites. Each of those IPsec Tunnels requires pair-wise
   periodic key refreshment. For a company with hundreds or thousands
   of locations, managing hundreds (or even thousands) of IPsec tunnels
   can be very processing intensive. That is why many Cloud operators
   only allow a limited number of (IPsec) tunnels and bandwidth to each
   customer.

   A solution like group key encryption [RFC4535] has been used to
   scale the IPsec key management. The group key management protocol
   documented in [RFC4535] outlines the relevant security risks for any
   group key management system in Section 3 (Security Considerations).
   While this particular protocol isn't being suggested, the drawbacks
   and risks of group key management are still relevant.

   [SDWAN-EDGE-DISCOVERY] leverages the peers communication polices on
   the SD-WAN controller and BGP Update messages to exchange IPsec
   Security Associations related parameters among peers without IKEv2
   point-to-point signaling or any other direct peer-to-peer session
   establishment messages.

Dunbar, et al.                                                [Page 15]
Internet-Draft     Net2Cloud Problems & Mitigations

5.2. CPEs Interconnection Over the Public Internet

   When enterprise CPEs are far away from each other, e.g., across
   country/continent boundaries, the performance of IPsec tunnels over
   the public Internet can be problematic and unpredictable. Even
   though there are many monitoring tools available to measure delay
   and various performance characteristics of the network, the
   measurement for paths over the Internet is passive and past
   measurements may not represent future performance.

   [MULTI-SEG-SDWAN] outlines some approaches for leveraging the Cloud
   backbone to connect enterprise CPEs across diverse geographical
   areas, eliminating the need for the Cloud GW to decrypt and re-
   encrypt traffic from the CPEs. A thorough examination of the
   security implications associated with this proposed method is
   necessary. Alternative encapsulations, like SRH (Segment Routing
   Header) or others, can be considered for interconnecting enterprise
   CPEs.

6. Requirements for Networks Connecting Cloud Data Centers

   To address the issues identified in this document, network solutions
   for connecting enterprises with their dynamic workloads or
   applications in Cloud DCs should satisfy the following requirements:
     - Should support scalable policy management for the traffic to
        and from the newly instantiated application instances at any
        Cloud DC location. The scalable policy management, even though
        out of the scope of this document, can include centralized
        policy repositories and API-driven automation.
     - Should allow enterprises to take advantage of the current
        state-of-the-art private VPN technologies, including the
        conventional circuit-based, MPLS-based VPNs, or IPsec-based
        VPNs (or any combination thereof) that run over the public
        Internet.
     - Should support scalable IPsec key management among all nodes
        involved in DC interconnect schemes.
     - Should support easy and fast, on-demand network connections to
        dynamic workloads and applications in Cloud DCs and easily
        reach these workloads when they migrate within or across data
        centers.

Dunbar, et al.                                                [Page 16]
Internet-Draft     Net2Cloud Problems & Mitigations

     - Should support traffic steering to distribute loads across
        regions/AZs based on performance/availability of workloads in
        addition to the network path conditions to the Cloud DCs.
     - Should support network traffic traceability, logging, and
        diagnostics.
     - Should support transit/spoke gateways interconnection
        scalability and consistent policy enforcement as workloads are
        increased/migrated. This requirement is mainly for the Cloud
        Aggregators or Cloud Service Brokers who provide managed
        services to enterprises over multiple Cloud service providers.

7. Security Considerations

   The security issues in terms of networking to Cloud DCs include:

     - Service instances in Cloud DCs are connected to users
        (enterprises) via Public IP ports which are exposed to the
        following security risks:

        a) Potential DDoS (Distributed Denial of Service) attack to the
        ports facing the untrusted network (e.g., the public internet),
        which may propagate to the cloud edge resources. To mitigate
        such security risk, it is necessary for the ports facing
        internet to enable Anti-DDoS features.

        b) Potential risk of augmenting the attack surface with inter-
        Cloud DC connection by means of identity spoofing, man-in-the-
        middle, eavesdropping or DDoS attacks. One example of
        mitigating such attacks is using DTLS to authenticate and
        encrypt MPLS-in-UDP encapsulation [RFC7510].

     - Potential attacks from service instances within the cloud. For
        example, data breaches, compromised credentials, and broken
        authentication, hacked interfaces and APIs, and account
        hijacking.

     - When IPsec tunnels established from enterprise on-premises CPEs
        are terminated at the Cloud DC gateway where the workloads or
        applications are hosted, traffic to/from an enterprise's
        workload can be exposed to others behind the data center

Dunbar, et al.                                                [Page 17]
Internet-Draft     Net2Cloud Problems & Mitigations

        gateway (e.g., exposed to other organizations that have
        workloads in the same data center).

        To ensure that traffic to/from workloads is not exposed to
        unwanted entities, IPsec tunnels may go all the way to the
        workload (servers, or VMs) within the DC.

     - Group key management [RFC4535] comes with security risks such
        as:  keys being used too long, single points of compromise (one
        compromise affects the whole group), key distribution
        vulnerabilities, key generation vulnerabilities, to name a few.

        [RFC4535] outlines the security risks in Section 3 (Security
        Considerations). While this specific protocol isn't being
        suggested the risks and vulnerabilities apply to any group key
        management system."

     - Striking a balance between scaling IPsec tunnel management
        outlined in this document and maintaining robust security is a
        delicate consideration. Simplifying the IPsec tunnel management
        to reduce management complexity for large SD-WAN networks might
        come with the inherent risk of decreased security. Careful
        consideration of the specific deployments, coupled with regular
        security assessments, is crucial to ensure the integrity and
        confidentiality of the transmitted data.

   The Cloud DC operator's security practices can affect the overall
   security posture and need to be evaluated by customers. Many Cloud
   operators offer monitoring services for data stored in Clouds, such
   as AWS CloudTrail, Azure Monitor, and many third-party monitoring
   tools to improve the visibility of data stored in Clouds.

   Solution drafts resulting from this work will address security
   concerns inherent to the solution(s), including both protocol
   aspects and the importance, for example, of securing workloads in
   cloud DCs and the use of secure interconnection mechanisms.

   A full security evaluation will be needed before [MULTI-SEG-SDWAN]
   can be recommended as a solution to some problems described in this
   document.

Dunbar, et al.                                                [Page 18]
Internet-Draft     Net2Cloud Problems & Mitigations

8. IANA Considerations

   This document requires no IANA actions.

9. References

9.1. Normative References

   [RFC3168] K. Ramakrishnan, et al, "The Addition of Explicit
             Congestion Notification (ECN) to IP", RFC3168, Sept. 2001.

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

   [RFC4486] E. Chen and V. Gillet, "Subcodes for BGP Cease
   Notification Message", RFC4486, April 2006.

   [RFC4535] H. Harney, et a, "GSAKMP: Group Secure Association Key
             Management Protocol", RFC4535, June 2006.

   [RFC4786] J. Abley and K. Lindqvist, "Operation of Anycast
             Services", RFC4786, Dec. 2006.

   [RFC5492] J. Scudder and R. Chandra, "Capabilities Advertisement
             with BGP-4", RFC5492, Feb. 2009.

   [RFC5880] D. Katz and D. Ward, "Bidirectional Forwarding Detection
             (BFD)", RFC5880, June 2010.

   [RFC6040] B. Briscoe, "Tunnelling of Explicit Congestion
             Notification", RFC6040, Nov 2010.

   [RFC6136] A. Sajassi and D. Mohan, "Layer 2 Virtual Private Network
             (L2VPN) Operations, Administration, and Maintenance (OAM)
             Requirements and Framework", RFC6136, March 2011.

   [RFC7606] E. Chen, et al "Revised Error Handling for BGP UPDATE
             Messages". Aug 2015.

Dunbar, et al.                                                [Page 19]
Internet-Draft     Net2Cloud Problems & Mitigations

   [RFC7432] A. Sajassi, et al "BGP MPLS-Based Ethernet VPN", RFC7432,
             Feb. 2015.

   [RFC7510] X. Xu, et al, "Encapsulating MPLS in UDP", RFC7510, April,
             2015.

   [RFC7938] P. Lapukhov, "Use of BGP for Routing in Large-Scale Data
             Centers", RFC7938, Aug. 2016.

   [RFC8490] R. Bellis, et al, "DNS Stateful Operations", RFC8490,
             March 2019.

   [RFC8765] T. Pusateri and S. Cheshire, "DNS Push Notifications",
             RFC8765, June 2020.

   [RFC8926] J. Gross and T. Sridhar, "Geneve: Generic Network
             Virtualization Encapsulation", RFC8926, Nov. 2020.

9.2. Informative References

   [RFC2735] B. Fox, et al "NHRP Support for Virtual Private networks".
             Dec. 1999.

   [RFC6071] S. Frankel and S. Krishnan, "IP Security (IPsec) and
             Internet Key Exchange (IKE) Document Roadmap", Feb 2011.

   [3GPP-5G-Edge] 3GPP TS 23.548 v18.1.1, "5G System Enhancements for
             Edge Computing", April 2023.

   [SDWAN-EDGE-DISCOVERY] L. Dunbar, S. Hares, R. Raszuk, K. Majumdar,
             G. Mishra, V. Kasiviswanathan, "BGP UPDATE for SD-WAN Edge
             Discovery", draft-ietf-idr-sdwan-edge-discovery-12, Oct
             2023.

   [AWS-NAT] NAT gateways - Amazon Virtual Private Cloud.

   [AWS-Cloud-WAN] Introducing AWS Cloud WAN (Preview) | Networking &
             Content Delivery (amazon.com).

   [Azure-SD-WAN] Architecture: Virtual WAN and SD-WAN connectivity -
             Azure Virtual WAN | Microsoft Learn.

Dunbar, et al.                                                [Page 20]
Internet-Draft     Net2Cloud Problems & Mitigations

   [NTW-AC] M. Boucadair, et al, "A Network YANG Data Model for
             Attachment Circuits", draft-ietf-opsawg-ntw-attachment-
             circuit-05, Feb. 2024.

   [DATAMODEL-BGP] M. Jethanandani, K. Patel, S. Hares, "YANG Model for
             Border Gateway Protocol (BGP-4)", draft-ietf-idr-bgp-
             model-17, July 2023.

   [Google-NAT] Cloud NAT overview  |  Google Cloud.

   [Int-tunnels] J. Touch and W Townsley, "IP Tunnels in the Internet
             Architecture", draft-ietf-intarea-tunnels-13.txt, March,
             2023.

   [MEF-70.1] MEF 70.1 SD-WAN Service Attributes and Service Framework.
             Nov. 2021.

   [METADATA-PATH] L. Dunbar, et al, "BGP Extension for 5G Edge Service
             Metadata" draft-ietf-idr-5g-edge-service-metadata-15, Feb.
             2024.

   [MULTI-SEG-SDWAN] K. Majumdar, et al, "Multi-segment SD-WAN via
             Cloud DCs", draft-dmk-rtgwg-multisegment-sdwan-07, Feb
             2024.

   [SVC-AC] M. Boucadair, et al. "YANG Data Models for 'Attachment
             Circuits'-as-a-Service (ACaaS)", draft-ietf-opsawg-teas-
             attachment-circuit-07, Feb. 2024.

   [Split-Horizon-DNS] K. Tirumaleswar, et al, "Establishing Local DNS
             Authority in Validated Split-Horizon Environments", draft-
             ietf-add-split-horizon-authority-07, Mar. 2023.

10. Acknowledgments

   Many thanks to Joel Halpern, Aseem Choudhary, Adrian Farrel, Alia
   Atlas, Chris Bowers, Mohamed Boucadair, Paul Vixie, Paul Ebersman,
   Timothy Morizot, Ignas Bagdonas, Donald Eastlake, Michael Huang, Liu
   Yuan Jiao, Katherine Zhao, and Jim Guichard for the discussion and
   contributions.

Dunbar, et al.                                                [Page 21]
Internet-Draft     Net2Cloud Problems & Mitigations

Authors' Addresses

   Linda Dunbar
   Futurewei
   Email: Linda.Dunbar@futurewei.com

   Andrew G. Malis
   Malis Consulting
   Email: agmalis@gmail.com

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

   Mehmet Toy
   Verizon
   One Verizon Way
   Basking Ridge, NJ 07920
   Email: mehmet.toy@verizon.com

   Kausik Majumdar
   Microsoft Azure
   kmajumdar@microsoft.com

Dunbar, et al.                                                [Page 22]