RTG Working Group L. Dunbar
Internet Draft Futurewei
Intended status: Informational A. Malis
Expires: March 15, 2026 Malis Consulting
C. Jacquenet
Orange
M. Toy
Verizon
K. Majumdar
Oracle
September 15, 2025
Networks to Cloud DCs: Challenges and Mitigation Practices
draft-ietf-rtgwg-net2cloud-problem-statement-43
Abstract
This document describes a set of network-related problems
enterprises face when interconnecting their branch offices with
dynamic workloads in third-party data centers (DCs) (a.k.a. Cloud
DCs). These challenges are particularly relevant to enterprises with
conventional VPN services that want to leverage those networks
(instead of altogether abandoning them).
The document also outlines various mitigation approaches, including
those already developed within the IETF. For challenges that do not
yet have established solutions, it identifies the IETF drafts that
have been proposed to address these issues. The intent is to provide
a cohesive view of problems and solution approaches that have been
documented or proposed within the IETF.
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."
Dunbar, et al. [Page 1]
Internet-Draft Net2Cloud Problems & Mitigations
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 March 15, 2026.
Copyright Notice
Copyright (c) 2025 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 Revised BSD License text as described in
Section 4.e of the Trust Legal Provisions and are provided without
warranty as described in the Revised BSD License.
Table of Contents
1. Introduction...................................................3
2. Conventions Used in This Document..............................3
3. Issues and Mitigation Methods of Connecting to Cloud DCs.......5
3.1. Increased BGP Peering Errors and Mitigation Methods.......5
3.2. Site Failures and Methods to Minimize Impacts.............7
3.3. Limitations of DNS-based Cloud DC Location Selection......8
3.4. Network Issues for 5G Edge Clouds and Mitigation Methods..8
3.5. DNS Practices for Hybrid Workloads........................9
3.6. NAT Practices for Accessing Cloud Services...............10
3.7. Cloud Discovery Practices................................11
4. Dynamic Connecting Enterprise Sites with Cloud DCs............11
4.1. Sites to Cloud DC........................................12
4.2. Inter-Cloud Connection...................................14
4.3. Extending Private VPNs to Hybrid Cloud DCs...............16
5. Methods to Scale IPsec Tunnels to Cloud DCs...................17
5.1. Scale IPsec Tunnels Management...........................17
5.2. CPEs Interconnection Over the Public Internet............18
6. Requirements for Networks Connecting Cloud Data Centers.......18
Dunbar, et al. [Page 2]
Internet-Draft Net2Cloud Problems & Mitigations
7. Security Considerations.......................................19
8. IANA Considerations...........................................21
9. References....................................................21
9.1. Normative References.....................................22
9.2. Informative References...................................23
10. Acknowledgments..............................................24
1. Introduction
Cloud data centers (DCs) provide scalable, on-demand services across
various geographic locations, enabling enterprises to deploy
applications and workloads closer to users for improved latency. The
dynamic nature of cloud workloads necessitates flexible networking
solutions to accommodate changes in service locations and
connectivity demands.
Cloud operators offer network functions such as virtual firewalls,
private cloud services, and virtual PBX systems. As a shared
infrastructure hosting multiple customers, Cloud DCs require
enterprises to establish robust connectivity solutions to integrate
existing VPNs with cloud networks.
This document examines networking challenges enterprises face when
connecting branch offices to Cloud DCs and explores mitigation
practices. While it references work from other standards development
organizations (SDOs), its primary focus remains within the IETF's
scope. Specifically, the document focuses on routing-related
challenges that have active IETF discussions and proposed solution
drafts, rather than attempting to address the entire problem space
of enterprise-cloud connectivity. Individual IETF solution drafts
address specific aspects of enterprise-cloud connectivity, but this
document unifies these elements to provide a comprehensive
perspective and emphasize the need for coordinated solutions.
References to IETF working groups and Internet Drafts are included
as examples to inform readers, without mandating the adoption of any
specific solution. Section 6 outlines high-level, solution-agnostic
requirements to guide future considerations in addressing these
challenges.
2. Conventions Used in This Document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
NOT","SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED",
Dunbar, et al. [Page 3]
Internet-Draft Net2Cloud Problems & Mitigations
"MAY", and "OPTIONAL" in this document are to be interpreted as
described in BCP14 [RFC2119] [RFC8174] when, and only when, they
appear in all capitals, as shown here.
Terms used in this document are described below.
Cloud DCs: Third party Data Centers that usually host applications
and workloads owned by different organizations or
tenants.
Hybrid 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
"edge" locations [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 [WIKI-IXP].
Private Cloud: The cloud infrastructure is provisioned for exclusive
use by a single organization comprising multiple
consumers (e.g., business units). It may be owned,
managed, and operated by the organization, a third
party, or some combination of them, and it may exist on
or off premises. (NIST Special Publication 800-145).
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-70.2]
VPC: A Virtual Private Cloud (VPC) is a secure, isolated
segment of a public cloud, where users can deploy and
manage resources such as virtual machines, databases,
and applications. VPCs offer the flexibility of using
Dunbar, et al. [Page 4]
Internet-Draft Net2Cloud Problems & Mitigations
the public cloud's infrastructure while providing more
control over networking and security.
3. Issues and Mitigation Methods of Connecting to Cloud DCs
This section identifies some high-level problems that can be
addressed using IETF technologies and ongoing standardization
efforts within the Routing area. Other Cloud DC related challenges,
such as managing Cloud spending or issues outside the Routing scope,
are out of the scope for this document.
3.1. Increased BGP Peering Errors and Mitigation Methods
Where conventional ISPs peer primarily with other ISPs and with a
limited number of VPN enterprise customers, public Cloud DCs
establish BGP sessions with a much larger and more diverse set of
enterprise customers. Many of these enterprises and application
providers are not experienced in managing complex BGP relationships,
which increases the likelihood of configuration errors, such as
capability mismatch, route leaks, missing Keepalives, and session
resets. Capability mismatch, in particular, can cause BGP sessions
not being adequately established. These issues are more acute for
Cloud DCs than they have been, though they also affect conventional
ISPs to a lesser degree.
BGP route convergence delays and security vulnerabilities, such as
BGP hijacking, remain significant concerns when connecting
enterprise networks to Cloud DCs. Route propagation policies and
peering configurations vary across cloud providers, requiring
enterprises to carefully design BGP session parameters to prevent
route leaks, session resets, and excessive route advertisements. The
use of BGP Route Reflectors, policy-based route filtering, and
automated session monitoring can help mitigate these risks and
improve BGP session stability in hybrid and multi-cloud
environments.
Here are the recommended mitigation practices:
- A Cloud GW typically establishes multiple eBGP sessions with
many clients. Each session is configured with a maximum number
of routes it can handle. To avoid exceeding this limit, which
could lead to the Cloud GW dropping routes, on-premises data
Dunbar, et al. [Page 5]
Internet-Draft Net2Cloud Problems & Mitigations
center gateways should simplify their route advertisements by
filtering unnecessary routes and using a default route instead.
This practice minimizes the volume of routing information
exchanged between on-premises data centers and Cloud DCs,
thereby preventing the unwanted dropping of routes when the
configured maximum for a client is exceeded, where appropriate
and when consistent with enterprise policy and Cloud DC
requirements.
- When a Cloud GW receives inbound routes exceeding the maximum
routes from a peer, the current practice is to generate out-of-
band alerts (e.g., Syslog entries) via the management system or
to terminate the BGP session (with a cease notification message
being sent per Section 4 of [RFC4486]). However, a more
operation-friendly approach would be for peers to reduce the
number of routes they are advertising. Therefore, it is worth
considering adding a "route threshold crossing" alert mechanism
to request peers to take action to reduce their advertised
routes, rather than their BGP sessions being terminated by
Cloud GW. While this mechanism is not available today and is
beyond the scope of this document, further discussion in the
IETF Inter-Domain Routing (IDR) Working Group is needed. Such
work could lead to the addition of new subcodes in RFC4486
Section 3 and corresponding descriptions in RFC4486 Section 4
to facilitate this more efficient approach.
- If a Cloud GW, a BGP speaker, receives from its BGP peer a
capability that it does not itself support or recognize, it
MUST ignore that capability, and the BGP session MUST NOT be
terminated per [RFC5492]. While ignoring unknown capabilities
prevents unnecessary session resets, cloud operators should
still monitor capability mismatches through logging or
management systems to avoid configuration ambiguities.
- 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, enterprise GWs need to establish tunnels
(e.g., IPsec) to the Cloud GWs to form an IP neighbor
relationship.
Dunbar, et al. [Page 6]
Internet-Draft Net2Cloud Problems & Mitigations
- 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,
where YANG interfaces are supported.
3.2. Site Failures and Methods to Minimize Impacts
In this document, a site refers to a subdivision within a Cloud Data
Center (Cloud DC), such as a building, a floor, a pod, or a server
rack.
Failures within a site can include capacity degradation or complete
out-of-service failure. Some examples of events that can trigger a
site failure are: 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 DC or between DCs.
As described in [RFC7938], a Cloud DC may not run IGP within its
domain, instead, it relies on internal methods to detect and report
faults, which differ from standardized protocols like BFD or IGP. In
the event of a site failure, while Cloud GW visible to clients
continues to operate normally, the failure remains undetected by
clients relying on BFD [RFC5880]. When BFD is not running within the
Cloud DC, the GW cannot simply extend or concatenate BFD sessions to
external peers.
When a site failure occurs, many services can be impacted. When the
impacted services' IP prefixes in a site are not well aggregated,
which is common, one single site failure can trigger multiple BGP
UPDATE messages. There are proposals, such as [METADATA-PATH], to
enhance BGP advertisements to reduce the number of messages
required.
[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. However, this alone is insufficient, as the routes at
the sites might not all be EVPN routes.
Dunbar, et al. [Page 7]
Internet-Draft Net2Cloud Problems & Mitigations
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 IP addresses 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, even
if the DNS TTL has expired.
- 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.
- No inherent awareness of proximity in the network (routing)
layer, resulting in suboptimal performance.
- Inflexible traffic control: The Local DNS resolver becomes the
unit of traffic management which requires DNS to receive
periodic updates of the network condition, which can be
operationally 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. However,
anycast optimizes based on routing reachability and may not reflect
real-time congestion or service load.
[METADATA-PATH] identifies metrics that can be utilized for the
ingress routers to make path steering decisions not only based on
the routing cost but also the running environment of the edge
services. This complements DNS-based approaches by shifting
decision-making to the routing layer.
[RFC8490] and [RFC8765] on stateful DNS can also help improve
performance by refreshing the cache and handling session idle
timeouts more effectively.
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 applications have low latency connections to the UEs
Dunbar, et al. [Page 8]
Internet-Draft Net2Cloud Problems & Mitigations
(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 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
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) While distances from the LDN Ingress router to server
instances in different edge clouds may vary slightly, the
overall service latency is significantly influenced by both
routing distance and capacity status at the edge cloud.
Therefore, a routing protocol solely based on the shortest
routing distance alone may not guarantee the lowest overall
latency. A more comprehensive approach that considers both
factors is essential for the routing protocol to achieve
service performance.
2) Due to user mobility, sources (UEs) can ingress from
different LDN Ingress routers, presenting a routing
challenge.
[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.
Dunbar, et al. [Page 9]
Internet-Draft Net2Cloud Problems & Mitigations
Each cloud operator has its own DNS to resolve resources within its
Cloud DCs and to well-known public domains. A cloud DNS service 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 using multiple cloud providers, 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. Some organizations use internal names
like those under a .internal top-level domain. However, .internal is
not an officially designated special-use domain name by IANA nor an
ICANN-approved Top-Level Domain. To avoid conflicts, enterprises
should use a globally unique, registered domain name, even for
internal resolution purposes. A globally unique name does not have
to be globally resolvable. An organization's domain can include
subdomains that are only resolvable within restricted zones, zones
that resolve differently depending on query origin, or zones that
resolve consistently for all queries [Split-Horizon-DNS].
Using globally unique names prevents collisions and simplifies
DNSSEC trust management, since registered domains can be chained to
the global DNSSEC trust anchor. Enterprises should therefore
consider using a registered FQDN from global DNS as the root for
both enterprise and 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. When
integrating multiple cloud environments or hybrid cloud
architectures, enterprises often face overlapping private IP address
spaces, requiring address translation techniques such as NAT.
Managing NAT policies across different cloud providers can introduce
additional complexity, particularly when ensuring consistent routing
and avoiding conflicts between overlapping RFC1918 address ranges.
By configuration, some private subnets can have NAT functionality to
reach out to external networks, while some private subnets are
internal to a Cloud DC only.
Dunbar, et al. [Page 10]
Internet-Draft Net2Cloud Problems & Mitigations
Different cloud operators support different levels of NAT
functionality. For example, in some environments a NAT gateway may
not support connections through private endpoints, VPN, direct
connections, or peering links [AWS-NAT]. In others, NAT services may
provide outbound connectivity to the Internet for instances without
public IP addresses, but not inbound NAT [Google-NAT]. These
variations mean that enterprises must carefully evaluate provider-
specific NAT features and limitations.
In addition to feature gaps across providers, NAT itself introduces
operational challenges. Address translation can obscure end-to-end
visibility, complicate troubleshooting, and make consistent policy
enforcement more difficult across multiple domains. NAT state
exhaustion and asymmetric routing can also lead to subtle service
disruptions.
For enterprises with applications running in different Cloud DCs,
NAT configurations must therefore be carefully coordinated across
Cloud DCs and on-premises DCs to ensure consistency, prevent
conflicts, and minimize operational complexity.
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,
as cloud operators can move the service instances from one place to
another. While geographic locations are usually exposed to
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,
particularly 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
Extension 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
Dunbar, et al. [Page 11]
Internet-Draft Net2Cloud Problems & Mitigations
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
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.
Each cloud provider enforces its own routing mechanisms, such as AWS
Transit Gateway, Azure Virtual WAN, and Google Cloud Dedicated
Interconnect. These vendor-specific architectures create additional
challenges for enterprises that require consistent routing policies
across multiple cloud environments.
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
Dunbar, et al. [Page 12]
Internet-Draft Net2Cloud Problems & Mitigations
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
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.
Dunbar, et al. [Page 13]
Internet-Draft Net2Cloud Problems & Mitigations
+------------------------+
| ,---. ,---. |
| (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. Inter-
cloud routing complexity arises from the lack of standardized
mechanisms for routing across multiple cloud providers. Each cloud
operator applies distinct routing policies, which can create
interoperability issues when establishing direct inter-cloud
connections. Enterprises may leverage third-party cloud service
brokers, SD-WAN overlays, or virtual routers instantiated in
different Cloud DCs to optimize traffic flow across cloud
environments.
Dunbar, et al. [Page 14]
Internet-Draft Net2Cloud Problems & Mitigations
Optimizing east-west traffic within and across Cloud DCs is critical
for modern workloads, particularly for applications with high inter-
service communication. Enterprises often rely on direct inter-VPC
peering, SD-WAN overlays, or cloud-native transit services (e.g.,
AWS Transit Gateway, Azure Virtual WAN) to improve performance and
reduce latency in multi-cloud and hybrid environments.
For example, when applications in AWS Cloud need to communicate with
applications in Azure, today's practice requires a third-party
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.
Dunbar, et al. [Page 15]
Internet-Draft Net2Cloud Problems & Mitigations
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.
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-
premise 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
Dunbar, et al. [Page 16]
Internet-Draft Net2Cloud Problems & Mitigations
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.
Managing hybrid cloud routing is further complicated by differences
in cloud provider routing architectures, making consistent policy
enforcement challenging. Enterprises often need to integrate SD-WAN
solutions or other overlay technologies to harmonize routing
behaviors across multiple cloud platforms.
To get around this problem, virtual routers in Cloud DCs can be used
to attach metadata (e.g., in the GENEVE header or IPv6 Extension
Header) to indicate the Geo-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
Dunbar, et al. [Page 17]
Internet-Draft Net2Cloud Problems & Mitigations
only allow a limited number of (IPsec) tunnels and bandwidth to each
customer.
A solution like group key management [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.
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) [RFC8754] 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
Dunbar, et al. [Page 18]
Internet-Draft Net2Cloud Problems & Mitigations
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.
- Should support traffic steering to distribute loads across
regions or Availability Zones 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
This document focuses on security challenges directly related to
networking and routing in enterprise-cloud connectivity, rather than
broader cloud security concerns such as encryption at rest, patch
management, and regulatory compliance. While those aspects are
important, they fall outside the scope of this document, which
specifically highlights network security risks, including BGP
security, DDoS mitigation, VPN scalability, and inter-cloud
connectivity risks. 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:
Dunbar, et al. [Page 19]
Internet-Draft Net2Cloud Problems & Mitigations
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. There are many Anti-DDoS
features to consider. Some examples include Rate Limiting,
Access Control Lists (ACLs), Deep Packet Inspection (DPI),
Blackholing and Sinkholing (which route malicious traffic to a
non-existent IP address or a system that safely absorbs or
analyzes the traffic), Traffic Scrubbing, and Geo-IP Blocking.
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
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.
- BGP security risks, including BGP hijacking and route leaks,
can lead to malicious traffic redirection. To mitigate these
risks, enterprises should implement BGP authentication (e.g.,
TCP MD5 or GTSM), RPKI for route validation, and strict
inbound/outbound route filtering. Additionally, session
security measures, such as RFC5492 for handling unsupported BGP
capabilities and RFC7606 for improved error handling, can
enhance routing stability and resilience.
Dunbar, et al. [Page 20]
Internet-Draft Net2Cloud Problems & Mitigations
- 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 [RFC4535] 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]
and [SDWAN-EDGE-DISCOVERY] can be recommended as a solution to some
problems described in this document.
8. IANA Considerations
This document requires no IANA actions.
9. References
Dunbar, et al. [Page 21]
Internet-Draft Net2Cloud Problems & Mitigations
9.1. Normative References
[RFC2119] S. Bradner, "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[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.
[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.
Dunbar, et al. [Page 22]
Internet-Draft Net2Cloud Problems & Mitigations
[RFC7938] P. Lapukhov, "Use of BGP for Routing in Large-Scale Data
Centers", RFC7938, Aug. 2016.
[RFC8174] B. Leiba, "Ambiguity of Uppercase vs Lowercase in RFC 2119
Key Words", RFC8174, May 2017.
[RFC8490] R. Bellis, et al, "DNS Stateful Operations", RFC8490,
March 2019.
[RFC8754] C. Filsfils, et al, "IPv6 Segment Routing Header (SRH)",
RFC8754, March 2020.
[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-20, Jan
2025.
[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 23]
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-12, July 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] https://cloud.google.com/nat/docs/overview
[HYBRID-CLOUD] https://cloud.google.com/learn/what-is-hybrid-cloud
[Int-tunnels] J. Touch and W Townsley, "IP Tunnels in the Internet
Architecture", draft-ietf-intarea-tunnels-13.txt, March
2023.
[MEF-70.2] MEF 70.2 SD-WAN Service Attributes and Service Framework.
Oct. 2023.
[METADATA-PATH] L. Dunbar, et al, "BGP Extension for 5G Edge Service
Metadata" draft-ietf-idr-5g-edge-service-metadata-26, Jan,
2025.
[MULTI-SEG-SDWAN] K. Majumdar, et al, "Multi-segment SD-WAN via
Cloud DCs", draft-ietf-rtgwg-multisegment-sdwan-01, June
2024.
[SVC-AC] M. Boucadair, et al. "YANG Data Models for 'Attachment
Circuits'-as-a-Service (ACaaS)", draft-ietf-opsawg-teas-
attachment-circuit-19, Jan 2025.
[Split-Horizon-DNS] K. Tirumaleswar, et al, "Establishing Local DNS
Authority in Validated Split-Horizon Environments", draft-
ietf-add-split-horizon-authority-14, June 2024.
[WIKI-IXP] https://en.wikipedia.org/wiki/Internet_exchange_point
10. Acknowledgments
Many thanks to Joel Halpern, Aseem Choudhary, Adrian Farrel, Alia
Atlas, Chris Bowers, Mohamed Boucadair, Paul Vixie, Paul Ebersman,
Dunbar, et al. [Page 24]
Internet-Draft Net2Cloud Problems & Mitigations
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 25]
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 26]
