Skip to content
All posts

Cisco ACI Design Guide: Fundamentals and Decisions

A Cisco ACI design starts with the fabric topology and controller cluster, then moves through tenant boundaries, routing, endpoint groups, security policy, workload connectivity, and external access. This Cisco ACI design guide explains those decisions in order, with the fundamentals you need to understand how they fit together.

In this guide:

What is Cisco ACI?

Cisco Application Centric Infrastructure (ACI) is a software-defined networking platform for data centers. It combines a fabric of Cisco Nexus switches with the Application Policy Infrastructure Controller (APIC). You define network and security policy through APIC, and the fabric applies that policy to workloads and their connections.

The change for a network engineer is the way the configuration is organized. Instead of starting with individual switch interfaces and access control lists, you describe the workloads, their network boundaries, and the communication they need. You still need to understand VLANs, routing, physical connectivity, and failure domains: ACI represents and implements those requirements through its policy model.

Design concern Traditional network approach Cisco ACI approach
Configuration Configure network devices and their interfaces. Define fabric and tenant policies through APIC.
Workload grouping Use constructs such as VLANs and subnets. Use bridge domains and endpoint groups to connect and group endpoints.
Traffic control Apply access control lists and other controls at selected points. Use contracts to define permitted communication between policy groups.
External routing Configure routed connections to the surrounding network. Define routed external connectivity through an L3Out.

You can practise these design decisions in CloudMyLab's hosted Cisco ACI learning labs. CloudMyLab Learning Labs use customizable physical topologies, giving you a place to work through the policy model before applying a design to production.

How the ACI policy model works

The policy model separates the logical network from the ports that attach workloads. These are the main objects to understand before designing a fabric:

  • Tenant: A container for policies and network objects, commonly used to organize an application, customer, or administrative boundary.
  • VRF: A routing context that holds IP connectivity and separates routing domains.
  • Bridge domain: A Layer 2 forwarding construct associated with a VRF. It can also contain gateway subnet definitions.
  • Endpoint group (EPG): A group of endpoints with common policy requirements. An application EPG is associated with a bridge domain.
  • Contract: A policy that specifies permitted traffic between groups that provide and consume it.
  • L3Out: A collection of objects that defines routed connectivity between a VRF and external networks.

For example, a web application might place web servers and application servers in separate EPGs. A contract permits the required web-to-application traffic. The EPGs connect to their bridge domains, and those bridge domains belong to a VRF. Access policies and EPG bindings connect the logical design to the physical or virtual workload interfaces.

This is a useful starting model, but it is not a rule that every subnet needs its own application or every EPG needs its own bridge domain. The right boundaries depend on routing, security, migration, and operations. For diagrams of the relationships, see Cisco ACI architecture diagrams.

Cisco ACI design decisions

Work from the physical fabric toward the workloads and their external dependencies. Record the reason for each boundary as well as the configuration: an operator should be able to trace a traffic flow from an endpoint to its destination and understand where policy applies.

1. Choose the fabric topology and leaf roles

A conventional ACI fabric uses a spine-leaf topology: leaf switches connect to the spine layer, and workloads and external networks connect to leaf switches. The fabric uses an overlay to carry endpoint traffic across the underlying network.

Start with the locations of servers, storage, network services, APICs, and external routers. Identify the leaf ports, link speeds, redundancy, and cabling each attachment needs. You may designate leaves for compute, external connectivity, or service appliances, but a separate pair for every function is not inherently required. Decide whether sharing leaves meets your capacity, availability, and operational requirements.

Document failure domains as well as normal traffic paths. For dual-connected workloads, check the supported attachment and virtual port-channel design. Validate switch models, optics, interfaces, and topology against Cisco's compatibility and scalability documentation for the software and hardware you plan to deploy. The Cisco ACI design guide explains the fabric infrastructure and attachment choices.

2. Design the APIC cluster

Use an APIC cluster for production management and resilience. Cisco describes a minimum cluster of three controllers; a single controller during initial fabric bootstrap is an exception, not the intended production end state. A lab or proof of concept may have different availability requirements.

Size the cluster for the intended fabric and verify the applicable scalability guidance. Place controllers and their leaf connections so that a local switch or site failure does not unnecessarily affect the cluster. Plan management access, time synchronization, name resolution, backups, and recovery before the fabric carries workloads.

APIC manages policy and the fabric, but it is not a transit hop for ordinary application traffic. Existing forwarding and the ability to make management changes are different availability questions, so test both in your failure scenarios. Cisco's APIC cluster documentation covers cluster management and resilience.

3. Define tenant boundaries

Choose tenants around meaningful ownership and policy boundaries. A tenant can group the VRFs, bridge domains, application profiles, EPGs, and contracts needed by a team or service. Separate tenants can make delegated administration and policy ownership easier to understand; they do not automatically replace a routing or security design.

Decide which services are shared and who owns them. ACI includes a common tenant for shared objects, but placing something there should be a deliberate choice: document how other tenants use it, which contracts are exposed, and who can change it. Avoid creating tenants solely to mirror every existing VLAN, or putting unrelated workloads together merely to reduce the number of objects.

Before you build, agree on naming, administrative roles, and the lifecycle of a tenant. For shared-service and cross-tenant design considerations, use the tenant networking sections of the Cisco ACI design guide.

4. Design VRFs and bridge domains

Use a VRF where you need a distinct routing context. Then define bridge domains around the Layer 2 and gateway requirements of the workloads within that context. A common starting point is a bridge domain for a subnet, but the design must account for the actual Layer 2 boundary, gateway placement, and migration needs.

For each bridge domain, establish where the default gateway lives and whether ACI will route for its endpoints. If ACI provides the gateway, define the appropriate subnet and verify its scope and advertisement requirements. If an external device remains the gateway during migration, check how traffic reaches it and which Layer 2 connectivity must remain in place.

Unicast routing and ARP flooding are design choices, not settings to copy without context. ACI can use endpoint learning and optimized forwarding, while some applications or migration scenarios depend on Layer 2 flooding behavior. Identify those dependencies, test endpoint discovery and failover, and consult Cisco's bridge domain design guidance before changing defaults.

5. Group endpoints with EPGs and ESGs

Choose groups based on the policy endpoints need. In a network-centric migration, an EPG can map closely to an existing VLAN and bridge domain. This can simplify the initial move, especially when the existing network boundaries and gateway behavior must be preserved.

In an application-centric design, you can divide workloads into groups such as web, application, and database tiers, then define the permitted flows between them. Multiple EPGs can share a bridge domain while remaining separate policy groups. You can use a mix of network-centric and application-centric approaches where that supports the migration.

Endpoint security groups (ESGs) offer another way to define security zones. An ESG operates within a VRF and is not tied to a single bridge domain. EPGs still provide endpoint-to-network association; ESGs can group endpoints for security policy across bridge domains in that VRF. Check platform support and the rules for using EPGs and ESGs together before adopting them. Cisco's ACI contract guide explains both models.

6. Define contracts and policy scope

Begin with a traffic matrix: source group, destination group, protocol, port, and the business reason for the flow. Translate approved flows into contracts and associate them with the appropriate provider and consumer groups. Test traffic that should be allowed and traffic that should be denied.

Be explicit about exceptions. Communication within an EPG is generally permitted by default, and policy enforcement can be affected by settings such as preferred groups, VRF policy enforcement, and intra-EPG isolation. Do not assume that creating EPGs alone produces the segmentation your security design requires.

Use vzAny when a contract relationship should apply broadly to eligible groups in a VRF, but review the resulting reach before using it for shared services. Preferred groups can allow communication among their members without individual contracts; that convenience also broadens connectivity. Contract scope controls where a contract can be applied, so choose it according to the intended boundary rather than making every contract broadly available. Check the Cisco ACI contract guide for the detailed behavior and exceptions.

7. Connect the policy model to interfaces

Access policies define how workloads attach to the fabric. Start with an inventory of leaf ports, connected devices, VLAN encapsulations, port channels, and the physical or virtual domains they require. Then build a consistent set of switch and interface profiles and policy groups for those attachments.

A VLAN pool identifies encapsulations available to a domain. An attachable access entity profile (AAEP) associates domains with interface policy groups. The EPG's domain association and, where appropriate, static path binding connect the workload's attachment to the logical policy. Each part must agree: an EPG can be correctly defined but still fail to carry traffic if its domain, VLAN, interface, or binding is wrong.

Use reusable policies where interfaces have the same requirements, and name them so an operator can trace a port back to its intended workload. Document vPC pairings and interface settings separately from the tenant traffic matrix. Cisco covers these relationships in the fabric access sections of its ACI design guide.

8. Plan the L3Out and external connectivity

An L3Out connects a VRF to external routed networks through border leaf interfaces. Design the physical links, routing protocol, route exchange, and external policy together. A working routing adjacency does not by itself mean an application is permitted to reach an external destination.

For each external connection, document the prefixes to advertise and receive, route filtering, default-route behavior, and failure paths. Define external EPGs and contracts for the traffic that must cross the fabric boundary. Check the interaction between external subnet classifications and contract policy, especially when broad prefixes overlap with more specific ones.

Test both directions of a flow and confirm the return route. Include dependencies such as DNS, authentication, monitoring, and management services in the traffic matrix. The L3Out sections of the Cisco ACI design guide cover routed connectivity and route-control considerations.

9. Choose Multi-Pod, Multi-Site, or Remote Leaf

Choose a distributed architecture based on the operational boundary and failure isolation you need, not just the distance between buildings. The interconnecting network and the services that span locations are part of the design.

Option When to consider it Design question
Multi-Pod You need a fabric distributed across pods with a shared APIC cluster and policy domain. Can the inter-pod network and shared operational boundary meet your availability and connectivity requirements?
Multi-Site You need separate ACI fabrics with coordinated policy across sites. Which policies and connectivity must span sites, and which failures and operations should remain local?
Remote Leaf You need to extend fabric connectivity to a remote location without deploying a separate fabric there. Does the remote connection meet Cisco's requirements for the intended services and failure behavior?

These choices have different implications for control, fault domains, endpoint mobility, and service appliances. Verify interconnect requirements, supported features, and recovery behavior for the planned hardware and software. Start with Cisco's ACI design guide and its Multi-Pod configuration guidance, then use the architecture-specific documentation for the option you select.

A practical Cisco ACI design example

Suppose you are moving a web application into an ACI fabric. It has web, application, and database workloads; users connect from an external network; and the application depends on shared DNS. The first deliverable is a traffic and ownership map, not an APIC configuration.

  1. Place the attachments. Identify the server-facing leaves, external routing connections, and any service appliances. Check redundant paths and port requirements.
  2. Choose the boundaries. Place the application objects in the tenant selected for its ownership model. Select a VRF and bridge domains that reflect the required routing and Layer 2 boundaries.
  3. Group the workloads. Create web, application, and database EPGs according to their communication requirements. Decide whether an ESG design would better express security policy across bridge domains.
  4. Describe permitted flows. Define contracts for the required web-to-application and application-to-database traffic. Include DNS and other shared dependencies.
  5. Attach endpoints and external networks. Associate the correct domains and bindings with the EPGs. Configure an L3Out and external policy for the required user and shared-service paths.
  6. Validate the result. Check endpoint learning, gateway reachability, routing, allowed flows, denied flows, and the return path. Repeat the tests after a link or device failure.

The result is a design you can explain and test. If the first migration preserves existing VLAN boundaries, document which segmentation is deferred and what must change before you introduce finer application policy.

Cisco ACI design checklist

Before approving a deployment, walk through the design with the teams responsible for networking, security, applications, and operations:

  • Fabric: Confirm supported hardware, cabling, attachment redundancy, capacity, and failure domains.
  • Management: Verify APIC cluster placement, management connectivity, DNS, time synchronization, backups, and recovery procedures.
  • Tenant and routing boundaries: Record ownership, VRFs, bridge domains, gateway locations, and route-sharing requirements.
  • Segmentation: Maintain a traffic matrix and review contracts, preferred groups, vzAny, and enforcement settings against it.
  • Attachments: Trace each workload from its port or virtual domain through the VLAN pool, AAEP, domain, and EPG association.
  • External connectivity: Check routing adjacencies, advertised and received prefixes, external EPGs, contracts, and return paths.
  • Operations: Define naming, change control, monitoring, fault response, and who owns shared policies.
  • Validation: Test permitted and denied traffic, endpoint learning, failover, and recovery using representative workloads.

Use Cisco's ACI best practices summary as a companion to the detailed design guide. Confirm each recommendation against your topology and use case rather than treating a checklist as a substitute for testing.

How to learn and test Cisco ACI

Start with the fabric and policy objects, then follow a packet through a small application. Create a tenant, VRF, bridge domain, and EPGs; attach endpoints; and test communication before and after applying a contract. Add an L3Out when you can explain the internal flow. Record the expected result before each change so you can distinguish a working configuration from an accidentally permissive one.

Next, practise troubleshooting. If an endpoint is missing, trace its interface, VLAN, domain association, and EPG binding. If an endpoint is learned but a flow fails, check the gateway, route, contract, filter, and return path. APIC faults and endpoint information help narrow the investigation, while packet captures and tests from the endpoints can confirm what traffic actually reaches them.

Automation is useful after the policy model is understood. APIC exposes an API, so teams can represent repeatable configurations in automation workflows and review changes before deployment. If you are evaluating Ansible for that work, start with how Ansible modules work and validate the ACI-specific modules and behavior against their documentation.

For broader preparation, see the network engineering certification and career guide and the guide to building a home or cloud network lab. When you are ready to practise ACI policy and design on a physical topology, explore CloudMyLab Cisco ACI learning labs.

FAQ

What does Cisco ACI stand for?

Cisco ACI stands for Cisco Application Centric Infrastructure. It is a data center networking platform that uses a policy model and the Application Policy Infrastructure Controller to manage a fabric of Cisco Nexus switches.

Can you practise Cisco ACI without buying hardware?

Yes. A hosted Cisco ACI learning lab lets you practise on a physical topology without purchasing the lab hardware yourself. Check the lab's available topology and access terms before choosing an exercise.

How long does it take to learn Cisco ACI?

The time depends on your networking experience, the depth you need, and your opportunity to practise. Start with the policy model and basic connectivity, then work through contracts, external routing, troubleshooting, and design validation.

How is Cisco ACI different from Cisco DNA Center?

Cisco ACI focuses on policy and networking for data center fabrics. Cisco DNA Center, known as Catalyst Center, focuses on managing and automating enterprise campus and branch networks. The platforms address different parts of the network.

What should a Cisco ACI design include?

An ACI design should cover fabric topology, APIC availability, tenant and routing boundaries, endpoint groups, contracts, access policies, external connectivity, operations, and failure testing. Distributed deployments also need an interconnect and site architecture.