Policy architecture
Policy architecture is a framework that
-
implements overlay network-wide policies called Cisco SD-WAN Validator policy or centralized policy
-
affects the flow of both control plane traffic and data plane traffic across the overlay network, and
-
supports traditional routing policies on Cisco IOS XE Catalyst SD-WAN devices for local routing protocol control.
Policy implementation details
The Cisco Catalyst SD-WAN policy architecture provides centralized policy configuration on a Cisco SD-WAN Controller. Cisco SD-WAN Controller policy affects control plane traffic through routing updates carried by Overlay Management Protocol (OMP) and used by the Cisco SD-WAN Controllers to determine the topology and status of the overlay network. It also affects data plane traffic that travels between the Cisco IOS XE Catalyst SD-WAN devices across the overlay network.
You can also create routing policies on the Cisco IOS XE Catalyst SD-WAN devices. These policies are traditional routing policies that are associated with routing protocol (BGP or OSPF) locally on the devices. You use them in the traditional sense for controlling BGP and OSPF, for example, to affect the exchange of route information, to set route attributes, and to influence path selection.
Centralized control policy architecture
A centralized control policy architecture is a network management framework that
-
enables centralized orchestration of network-wide routing decisions by a central authority instead of hop-by-hop implementation
-
allows influence over network routes advertised by Cisco SD-WAN Controllers through policies provisioned centrally
-
ensures routing policy results are distributed to devices using the Overlay Management Protocol (OMP) while maintaining centralized policy configuration.
Architecture components and implementation
In the Cisco IOS XE Catalyst SD-WAN network architecture, centralized control policy is handled by the Cisco SD-WAN Controller, which effectively is the routing engine of the network. The Cisco SD-WAN Controller is the centralized manager of network-wide routes, maintaining a primary route table for these routes. The Cisco SD-WAN Controller builds its route table based on the route information advertised by the Cisco IOS XE Catalyst SD-WAN devices in its domain, using these routes to discover the network topology and to determine the best paths to network destinations.
The Cisco SD-WAN Controller distributes route information from its route table to the devices in its domain which in turn use these routes to forward data traffic through the network. The result of this architecture is that networking-wide routing decisions and routing policy are orchestrated by a central authority instead of being implemented hop by hop, by the devices in the network.
Centralized control policy allows you to influence the network routes advertised by the Cisco SD-WAN Controllers. This type of policy, which is provisioned centrally on the Cisco SD-WAN Controller, affects both the route information that the Cisco SD-WAN Controller stores in its primary route table and the route information that it distributes to the devices.
Centralized control policy is provisioned and applied only on the Cisco SD-WAN Controller. The control policy configuration itself is never pushed to devices in the overlay network. What is pushed to the devices, using the Overlay Management Protocol (OMP), are the results of the control policy, which the devices then install in their local route tables and use for forwarding data traffic. This design means that the distribution of network-wide routes is always administered centrally, using policies designed by network administrators. These policies are always implemented by centralized Cisco SD-WAN Controllers, which are responsible for orchestrating the routing decisions in the Cisco IOS XE Catalyst SD-WAN overlay network.
Within a network domain, the network topology map on all Cisco SD-WAN Controllers must be synchronized. To support this, you must configure identical policies on all the Cisco SD-WAN Controllers in the domain.
All centralized control plane traffic, including route information, is carried by OMP peering sessions that run within the secure, permanent DTLS connections between devices and the Cisco SD-WAN Controllers in their domain. The end points of an OMP peering session are identified by the system IDs of the devices, and the peering sessions carry the site ID, which identifies the site in which the device is located. A DTLS connection and the OMP session running over it remain active as long as the two peers are operational.
Control policy can be applied both inbound, to the route advertisements that the Cisco SD-WAN Controller receives from the devices, and outbound, to advertisements that it sends to them. Inbound policy controls which routes and route information are installed in the local routing database on the Cisco SD-WAN Controller, and whether this information is installed as-is or is modified. Outbound control policy is applied after a route is retrieved from the routing database, but before a Cisco SD-WAN Controller advertises it, and affects whether the route information is advertised as-is or is modified.
Route types
Route types are OMP route categorizations that
-
carry prefix information that the devices learn from routing protocols
-
identify transport locations and their properties, and
-
identify available network services on local-site networks.
Route type details
The Cisco SD-WAN Controller learns the network topology from OMP routes, which are Cisco IOS XE Catalyst SD-WAN-specific routes carried by OMP. There are three types of OMP routes:
-
Cisco IOS XE Catalyst SD-WAN OMP routes—These routes carry prefix information that the devices learn from the routing protocols running on its local network, including routes learned from BGP and OSPF, as well as direct, connected, and static routes. OMP advertises OMP routes to the Cisco SD-WAN Controller by means of an OMP route SAFI (Subsequent Address Family Identifier). These routes are commonly simply called OMP routes.
-
TLOC routes—These routes carry properties associated with transport locations, which are the physical points at which the devices connect to the WAN or the transport network. Properties that identify a TLOC include the IP address of the WAN interface and a color that identifies a particular traffic flow. OMP advertises TLOC routes using a TLOC SAFI.
-
Service routes—These routes identify network services, such as firewalls and IDPs, that are available on the local-site network to which the devices are connected. OMP advertises these routes using a service SAFI.
The difference in these three types of routes can be viewed by using the various show sdwan OMP operational commands when you are logged in to the CLI on a Cisco SD-WAN Controller or a Cisco IOS XE Catalyst SD-WAN device. The show sdwan OMP routes command displays information sorted by prefix, the show sdwan OMP services command displays route information sorted by service, and the show sdwan OMP tlocs command sorts route information by TLOC.
Default behavior without centralized control policy
Default behavior without centralized control policy is a network state that
-
operates when no centralized control policy is provisioned on the Cisco SD-WAN Controller
-
enables unrestricted sharing of all route information across controllers and devices within a domain, and
-
results in automatic redistribution of OMP, TLOC, and service routes throughout the network.
Route advertisement and redistribution behavior
The default behavior produces the following route advertisement and redistribution patterns within a domain:
-
All Cisco IOS XE Catalyst SD-WAN devices redistribute all the route-related prefixes that they learn from their site-local network to the Cisco SD-WAN Controller. This route information is carried by OMP route advertisements that are sent over the DTLS connection between the devices and the Cisco SD-WAN Controller. If a domain contains multiple Cisco SD-WAN Controllers, the devices send all OMP route advertisements to all the controllers.
-
All the devices send all TLOC routes to the Cisco SD-WAN Controller or controllers in their domain, using OMP.
-
All the devices send all service routes to advertise any network services, such as firewalls and IDPs, that are available at the local site where the device is located. Again, these are carried by OMP.
-
The Cisco SD-WAN Controller accepts all the OMP, TLOC, and service routes that it receives from all the devices in its domain, storing the information in its route table. The Cisco SD-WAN Controller tracks which OMP routes, TLOCs, and services belong to which VPNs. The Cisco SD-WAN Controller uses all the routes to develop a topology map of the network and to determine routing paths for data traffic through the overlay network.
-
The Cisco SD-WAN Controller redistributes all information learned from the OMP, TLOC, and service routes in a particular VPN to all the devices in the same VPN.
-
The devices regularly send route updates to the Cisco SD-WAN Controller.
-
The Cisco SD-WAN Controller recalculates routing paths, updates its route table, and advertises new and changed routing information to all the devices.
Behavior changes with centralized control policy
A centralized control policy is a network management mechanism that
-
filters, modifies, or rejects route information before distribution to Cisco IOS XE Catalyst SD-WAN devices in a domain
-
controls route information stored in the Cisco Catalyst SD-WAN Controller's route table or advertised by the Cisco Catalyst SD-WAN Controller, and
-
applies to specific sites in the overlay network in either inbound or outbound direction with respect to the Cisco Catalyst SD-WAN Controller.
Control policy direction behaviors
All provisioning of centralized control policy is done on the Cisco Catalyst SD-WAN Controller.
Control policy behavior depends on the direction of application:
-
Inbound direction: Filters or modifies the routes being advertised by the Cisco IOS XE Catalyst SD-WAN device before they are placed in the route table on the Cisco Catalyst SD-WAN Controller. Routes are either accepted or rejected as the first step. Accepted routes are installed in the route table on the Cisco Catalyst SD-WAN Controller either as received or as modified by the control policy. Routes that are rejected by a control policy are silently discarded.
-
Outbound direction: Filters or modifies the routes that the Cisco Catalyst SD-WAN Controller redistributes to the Cisco IOS XE Catalyst SD-WAN devices. Routes are either accepted or rejected as the first step. For accepted routes, centralized control policy can modify the routes before they are distributed by the Cisco Catalyst SD-WAN Controller. Routes that are rejected by an outbound policy are not advertised.
VPN membership policy is a second type of centralized data policy that controls whether a Cisco IOS XE Catalyst SD-WAN device can participate in a particular VPN. VPN membership policy defines which VPNs of a device is allowed and which is not allowed to receive routes from.
VPN membership policy can be centralized because it affects only the packet headers and has no impact on the choice of interface that a Cisco IOS XE Catalyst SD-WAN device uses to transmit traffic. If, because of a VPN membership policy, a device is not allowed to receive routes for a particular VPN, the Cisco Catalyst SD-WAN Controller never forwards those routes to that driver.
Examples of modifying traffic flow with centralized control policy
Examples of modifying traffic flow with centralized control policy are practical scenarios that demonstrate how you can use centralized control policies to modify the flow of data traffic through the overlay network.
Basic examples of traffic flow modification
This section provides some basic examples of how you can use centralized control policies to modify the flow of data traffic through the overlay network.
Arbitrary topologies
-
uses control policies to create specific network structures like hub-and-spoke models,
-
minimizes CPU resource overhead by reducing the number of direct IPsec tunnel connections, and
-
enables scalable network architectures for large deployments with hundreds or thousands of branches.
Hub-and-spoke topology implementation
When data traffic is exchanged between two Cisco IOS XE Catalyst SD-WAN devices, if you have provisioned no control policy, the two devices establish an IPsec tunnel between them and the data traffic flows directly from one device to the next. For a network with only two devices or with just a small number of devices, establishing connections between each pair of devices is generally not been an issue. However, such a solution does not scale. In a network with hundreds or even thousands of branches, establishing a full mesh of IPsec tunnels tax the CPU resources of each device.
One way to minimize this overhead is to create a hub-and-spoke type of topology in which one of the devices acts as a hub site that receives the data traffic from all the spoke, or branch, devices and then redirects the traffic to the proper destination. This example shows one of the ways to create such a hub-and-spoke topology, which is to create a control policy that changes the address of the TLOC associated with the destination.
West-East branch hub-and-spoke configuration
The figure illustrates how such a policy might work. The topology has two branch locations, West and East. When no control policy is provisioned, these two devices exchange data traffic with each other directly by creating an IPsec tunnel between them (shown by the red line). Here, the route table on the Device West contains a route to Device East with a destination TLOC of 203.0.113.1, color gold (which we write as the tuple {192.0.2.1, gold}), and Device East route table has a route to the West branch with a destination TLOC of {203.0.113.1, gold}.
To set up a hub-and-spoke–type topology here, we provision a control policy that causes the West and East devices to send all data packets destined for the other device to the hub device. (Remember that because control policy is always centralized, you provision it on the Cisco Catalyst SD-WAN Controller.) On the Device West, the policy simply changes the destination TLOC from {203.0.113.1, gold} to {209.165.200.225, gold}, which is the TLOC of the hub device, and on the Device East, the policy changes the destination TLOC from {192.0.2.1, gold} to the hub's TLOC, {209.165.200.225, gold}. If there were other branch sites on the west and east sides of the network that exchange data traffic, you could apply these same two control policies to have them redirect all their data traffic through the hub.
Traffic engineering
Traffic engineering is a control policy capability that
-
allows you to design and provision traffic flow through the network using TLOC preferences
-
enables path tracking to monitor end-to-end path availability between source and destination devices, and
-
provides service chaining functionality by redirecting packets through network services such as firewalls.
Traffic engineering implementation methods
Control policy allows you to design and provision traffic engineering. In a simple case, suppose that you have two devices acting as hub devices. If you want data traffic destined to a branch Cisco IOS XE Catalyst SD-WAN device to always transit through one of the hub devices, set the TLOC preference value to favor the desired hub device.
The figure shows that Site ID 100 has two hub devices, one that serves the West side of the network and a second that serves the East side. Data traffic from the Device West must be handled by the Device West hub, and similarly, data traffic from the Device East branch must go through the Device East hub.
To engineer this traffic flow, you provision two control policies, one for Site ID 1, where the Device West device is located, and a second one for Site ID 2. The control policy for Site ID 1 changes the TLOC for traffic destined to the Device East to {209.165.200.225, gold}, and the control policy for Site ID 2 changes the TLOC for traffic destined for Site ID 1 to {198.51.100.1, gold}. One additional effect of this traffic engineering policy is that it load-balances the traffic traveling through the two hub devices.
With such a traffic engineering policy, a route from the source device to the destination device is installed in the local route table, and traffic is sent to the destination regardless of whether the path between the source and destination devices is available. Enabling end-to-end tracking of the path to the ultimate destination allows the Cisco Catalyst SD-WAN Controller to monitor the path from the source to the destination, and to inform the source device when that path is not available. The source device can then modify or remove the path from its route table.
![]() Note |
Ensure that the Traffic Engineering (TE) service is enabled on the intermediate hub to utilize the traffic engineering functionalities. |
The figure Traffic Engineering 2 illustrates end-to-end path tracking. It shows that traffic from Device-A that is destined for Device-D first goes to an intermediate device, Device-B, perhaps because this intermediate device provides a service, such as a firewall. (You configure this traffic engineering with a centralized control policy that is applied to Device-A, at Site 1.) Then Device-B, which has a direct path to the ultimate destination, forwards the traffic to Device-D. So, in this example, the end-to-end path between Device-A and Device-D comprises two tunnels, one between Device-A and Device-B, and the second between Device-B and Device-D. The Cisco Catalyst SD-WAN Controller tracks this end-to-end path, and it notifies Device-A if the portion of the path between Device-B and Device-D becomes unavailable.
As part of end-to-end path tracking, you can specify how to forwarded traffic from the source to the ultimate destination using an intermediate device. The default method is strict forwarding, where traffic is always sent from Device-A to Device-B, regardless of whether Device-B has a direct path to Device-D or whether the tunnel between Device-B and Device-D is up. More flexible methods forward some or all traffic directly from Device-A to Device-D. You can also set up a second intermediate device to provide a redundant path with the first intermediate device is unreachable and use an ECMP method to forward traffic between the two. The figure Traffic Engineering3 adds Device-C as a redundant intermediate device.
Centralized control policy, which you configure on Cisco Catalyst SD-WAN Controllers, affects routing policy based on information in OMP routes and OMP TLOCs.
In domains with multiple Cisco Catalyst SD-WAN Controllers, all the controllers must have the same centralized control policy configuration to ensure that routing within the overlay network remains stable and predictable.
Centralized policy-based configuration on prefixes and IP headers
A centralized data policy based on source and destination prefixes and on headers in IP packets is a series of numbered sequences that
-
consists of ordered match-action pairs that are evaluated from lowest sequence number to highest sequence number
-
takes the associated action when a packet matches one of the match conditions and stops policy evaluation on that packet, and
-
drops and discards packets by default if no parameters in any of the sequences match.
Configuration components
The following figure illustrates the configuration components for a centralized data policy:

Feedback