Overview

This guide contains the maximum verified scalability limits for Cisco Application Centric Infrastructure (Cisco ACI) parameters in the following releases:

  • Cisco Application Policy Infrastructure Controller (Cisco APIC), Release 6.0(3)

  • Cisco Nexus 9000 Series ACI-Mode Switches, Release 16.0(3)

These values are based on a profile where each feature was scaled to the numbers specified in the tables. These numbers do not represent the theoretically possible Cisco ACI fabric scale.


Note

The verified scalability limits for Cisco Multi-Site previously included as part of this guide are now listed in the Cisco Nexus Dashboard Orchestrator (NDO) release-specific documents available at the following URL: https://www.cisco.com/c/en/us/support/cloud-systems-management/multi-site-orchestrator/products-device-support-tables-list.html.

The verified scalability limits for Cisco Cloud APIC previously included as part of this guide are now listed in the Cloud APIC release-specific documents available at the following URL: https://www.cisco.com/c/en/us/support/cloud-systems-management/cloud-application-policy-infrastructure-controller/products-tech-notes-list.html.

New and Changed Information

These changes have been made to this document since the initial release:

Date

Changes

October 21, 2024

Updated General Scalability L4-L7 Devices topic to 1,200 physical or virtual devices and Layer 4 to Layer 7 topic by changing configurable options to add "concrete devices."

October 8, 2024 Updated High Dual Stack profile values in Fabric Topology section.

August 27, 2024

Updated NetFlow Scale information.
August 20, 2024 Updated VRFs Per Leaf Scale value and the External EPGs values in Fabric Topology section.

August 19, 2024

Updated "Per Fabric Scale" values of the IP SLA probes in the Fabric Topology section.

June 28, 2024

Updated General Scalability section with a Note about fabric having 32 GB of RAM. Updated Fabric Topology content with notes about EP scale, fabric-wide, fabric having 32 GB of RAM.

Updated Fabric Topology content with updated scale guidelines for the number of Consumers and Providers for the same contract and Number of VLAN encapsulations per EPG.

November 2, 2023

Added the Nutanix table to the VMM Scalability Limits.
August 9, 2023 First release of this document.

General Scalability Limits

  • L2 Fabric: L2 Fabric in this document refers to an ACI fabric that contains only BDs with Scaled L2 Only mode (formerly known as Legacy mode). See Bridging > Bridge Domain Options > Scaled L2 Only Mode - Legacy Mode in APIC Layer 2 Configuration Guide for details about Scaled L2 Only mode.

  • L3 Fabric: The ACI L3 fabric solution provides a feature-rich highly scalable solution for public cloud and large enterprise. With this design, almost all supported features are deployed at the same time and are tested as a solution. The scalability numbers listed in this section are multi-dimensional scalability numbers. The fabric scalability numbers represent the overall number of objects created on the fabric. The per-leaf scale numbers are the objects created and presented on an individual leaf switch. The fabric level scalability numbers represent APIC cluster scalability and the tested upper limits. Some of the per-leaf scalability numbers are subject to hardware restrictions. The per-leaf scalability numbers are the maximum limits tested and supported by leaf switch hardware. This does not necessarily mean that every leaf switch in the fabric was tested with maximum scale numbers.

  • Stretched Fabric: Stretched fabric allows multiple fabrics (up to 3) distributed in multiple locations to be connected as a single fabric with a single management domain. The scale for the entire stretched fabric remains the same as for a single site fabric. For example a L3 stretched fabric will support up to 400 leaf switches total which is the maximum number of leaf switches supported on a single site fabric. Parameters only relevant to stretched fabric are mentioned in the tables below.

  • Multi-Pod: Multi-Pod enables provisioning a more fault-tolerant fabric comprised of multiple Pods with isolated control plane protocols. Also, Multi-Pod provides more flexibility with regard to the full mesh cabling between leaf and spine switches. For example, if leaf switches are spread across different floors or different buildings, Multi-Pod enables provisioning multiple Pods per floor or building and providing connectivity between Pods through spine switches.

    Multi-Pod uses a single APIC cluster for all the Pods; all the Pods act as a single fabric. Individual APIC controllers are placed across the Pods but they are all part of a single APIC cluster.

  • Multi-Site: Multi-Site is the architecture interconnecting and extending the policy domain across multiple APIC cluster domains. As such, Multi-Site could also be named as Multi-Fabric, since interconnects separate Availability Zones (Fabrics) and managed by an independent APIC controller cluster. A Cisco Nexus Dashboard Orchestrator (NDO) is part of the architecture and is used to communicate with the different APIC domains to simplify the management of the architecture and the definition of inter-site policies.

Leaf Switches and Ports

The maximum number of leaf switches is 400 per Pod and 500 total in Multi-Pod fabric. The maximum number of physical ports is 24,000 per fabric. The maximum number of remote leaf (RL) switches is 200 per fabric, with total number of BDs deployed on all remote leaf switches in the fabric not exceeding 60,000. The total number of BDs on all RLs is equal to the sum of BDs on each RL.

If Remote Leaf Pod Redundancy policy is enabled, we recommended that you disable the Pre-emption flag in the APIC for all scaled up RL deployments. In other words, you must wait for BGP CPU utilization to fall under 50% on all spine switches before you initiate pre-emption.

Breakout Ports

The N9K-C9336C-FX2 switch supports up to 34 breakout ports in both 10G or 25G mode.

General Scalability Limits


Note

For large fabrics, we recommend that all spines in the fabric have 32 GB of RAM.
Table 1. Fabric Scale Limits Per Cluster Size

Configurable Options

Default Fabric

Medium Fabric

Large Fabric

Number of APIC nodes

3

4

5 or 6

7

Number of leaf switches

85

200

300

500

Number of leaf switches per Pod

85

200

200

400

Number of tier-2 leaf switches per Pod in Multi-Tier topology

Note

 

The total number of leaf switches from all tiers must not exceed the "Number of leaf switches" listed above.

80

100

125

125

Number of Pods

6

6

25

25

Number of spine switches in a Multi-Pod fabric

24

24

50

50

Number of tenants

1,000

1,000

3,000

3,000

Number of Layer 3 (L3) contexts (VRFs)

1,000

1,000

10,000

10,000

Number of L3Outs

2,400

2,400

10,000

10,000

Number of external EPGs across all BLs

This is calculated as a product of (Number of external EPGs)*(Number of border leaf switches for the L3Out).

For example, the following combination adds up to a total of 2000 external EPGs in the fabric (250 external EPGs * 2 border leaf switches * 4 L3Outs):

  • 250 External EPGs in L3Out1 on leaf1 and leaf2

  • 250 External EPGs in L3Out2 on leaf1 and leaf2

  • 250 External EPGs in L3Out3 on leaf3 and leaf4

  • 250 External EPGs in L3Out4 on leaf3 and leaf4

4,000

4,000

10,000

10,000
Table 2. General Scalability Limits Per Fabric

Configurable Options

Scale Limits

Number of spine switches per Pod

6

Number of FEXs

650

(maximum of 20 FEXs and 576 ports per leaf)

Number of contracts

10,000

Number of contract filters

10,000

Number of endpoint groups (EPGs)

15,000

(21,000 for L2 fabric)

Number of EPGs per tenant

General limits:

  • Single-tenant fabrics: 4,000

  • Multi-tenant fabrics: 500

Or one of the following two specific use cases within the same fabric (the EPGs must be deployed on local leaf switches only, not on remote leaf switches):

  • Use case 1:

    • Up to 10 tenants that have up to 700 EPGs per tenant, with the EPGs distributed across up to 100 leaf switches

  • Use case 2:

    • 1 tenant with up to 1,400 EPGs deployed on up to 100 leaf switches

      For example, tenant1 with EPG1-1400 on leaf1-100

    • 1 tenant with up to 800 EPGs deployed on a different set of up to 20 leaf switches

      For example, tenant2 with EPG1401-2200 on leaf101-120

    • 2 tenants with up to 800 EPGs per tenant deployed on a different set of up 20 leaf switches

      For example, tenant3 with EPG2201-3000 and tenant4 with EPG 3001-3800 on leaf121-140

Number of bridge domains (BDs)

15,000

(21,000 for L2 fabric)

Number of vCenters

200 VDS

Number of Service Chains

1,000

Number of L4-L7 concrete devices

1,200 physical or virtual devices (1,200 maximum in total per fabric)

Number of ESXi hosts - VDS

3,200

Number of VMs

Depends on server scale

Number of configuration zones per fabric

30

L3 EVPN services over fabric WAN - GOLF (with and without OpFlex)

1,000 VRFs

60,000 routes in a fabric

Number of Routes in Overlay-1 VRF

1,000

Floating L3Out

6 anchor nodes

32 non-anchor nodes

Multiple Fabric Options Scalability Limits

Stretched Fabric

Configurable Options

Per Fabric Scale

Number of fabrics that can be a stretched fabric

3

Number of Route Reflectors

6

Multi-Pod

Configurable Options

Per Fabric Scale

Number of Pods

25

Number of leaf switches per Pod

400

Number of leaf switches overall

500

Number of Route Reflectors for L3Out

50

Number of External Route Reflectors between Pods

  • For 1-3 Pods: Up to 3 external route reflectors

    We recommend full mesh for external BGP peers instead of using external route reflectors when possible

  • For 4 or more Pods: Up to 4 external route reflectors

    We recommend using external route reflectors instead of full mesh

    We recommend that the external route reflectors are distributed across Pods so that in case of any failure there are always at least two Pods with external route reflectors still reachable

Cisco Multi-Site Scalability Limits

Cisco Nexus Dashboard Orchestrator (NDO) does not require a specific version of APIC to be running in all sites. The APIC clusters in each site as well as the NDO itself can be upgraded independently of each other and run in mixed operation mode as long as each fabric is running APIC, Release 3.2(6) or later.

As such, the verified scalability limits for your specific Cisco Nexus Dashboard Orchestrator release are now available at the following URL: https://www.cisco.com/c/en/us/support/cloud-systems-management/multi-site-orchestrator/products-device-support-tables-list.html.


Note

Each site managed by the Cisco Nexus Dashboard Orchestrator must still adhere to the scalability limits specific to that site's APIC Release. For a complete list of all Verified Scalability Guides, see https://www.cisco.com/c/en/us/support/cloud-systems-management/application-policy-infrastructure-controller-apic/tsd-products-support-series-home.html#Verified_Scalability_Guides

Fabric Topology, SPAN, Tenants, Contexts (VRFs), Equal Cost Multipath (ECMP), External EPGs, Bridge Domains, Endpoints, and Contracts Scalability Limits

This content shows the mapping of the "Application Leaf Engine (ALE) and Leaf Spine Engine (LSE) type" to the corresponding leaf switches. The information is helpful to determine which leaf switch is affected when we use the terms LSE or LSE2 in the remaining sections.


Note

The switches are listed as LSE or LSE2 for scalability purposes only. Check specific feature documentation for the full list of supported devices.

LSE Type

ACI-Supported Leaf Switches

LSE

  • N9K-C93108TC-EX

  • N9K-C93108TC-EX-24

  • N9K-C93180YC-EX

  • N9K-C93180YC-EX-24

  • N9K-C93180LC-EX

  • N9K-C9336C-FX2

  • N9K-C93216TC-FX2

  • N9K-C93240YC-FX2

  • N9K-C93360YC-FX2

  • N9K-C9336C-FX2-E

  • N9K-C9364D-GX2A

  • N9K-C9348D-GX2A

  • N9K-C9400-SW-GX2A

LSE2

  • N9K-C93108TC-FX

  • N9K-C93108TC-FX-24

  • N9K-C93180YC-FX

  • N9K-C93180YC-FX-24

  • N9K-C9348GC-FXP

  • N9K-C93600CD-GX

  • N9K-C9364C-GX

  • N9K-C9316D-GX

  • N9K-C9332D-GX2B

  • N9K-C93180YC-FX3

  • N9K-C93108TC-FX3P

  • N9K-C9358GY-FXP with 24GB of RAM

  • N9K-C93180YC-FX3H

  • N9K-C93108TC-FX3H

Note

  • The High Policy, Multicast-Heavy, and High IPv4 EP Scale profiles are not supported on FXP switches.

  • Full scale support for High Policy, Multicast-Heavy, and High IPv4 EP Scale profiles requires LSE2 with 32 GB of RAM.

  • High IPv4 EP Scale—We recommend using this profile only for the ACI border leaf (BL) switches in Multi-Domain (ACI-SDA) Integration. It provides enhanced IPv4 EP and LPM scales specifically for these BLs and has specific hardware requirements.

  • For maximum EP scale, fabric-wide, we recommend that all spines in the fabric have 32 GB of RAM.

  • For full scale support of Maximum LPM scale profile, we recommend that all spines in the fabric have 32 GB of RAM.

For more details on Forwarding Scale Profiles and the list of supported devices, refer to Cisco APIC Forwarding Scale Profiles at this url: https://www.cisco.com/c/en/us/td/docs/switches/datacenter/aci/apic/sw/all/forwarding-scale-profiles/cisco-apic-forwarding-scale-profiles.html

Fabric Topology

Configurable Options

Per Leaf Scale

Per Fabric Scale

Number of PCs, vPCs

320 (with FEX HIF)

N/A

Number of encapsulations per access port, PC, vPC (non-FEX HIF)

3,000

N/A

Number of encapsulations per FEX HIF, PC, vPC

100

N/A

Number of encapsulations per FEX

1,400

N/A

Number of member links per PC, vPC*

*vPC total ports = 32, 16 per leaf

16

N/A

Number of ports x VLANs (global scope and no FEX HIF)

64,000

168,000 (when using legacy BD mode)

N/A

Number of ports x VLANs (FEX HIFs and/or local scope)

10,000

N/A

Number of static port bindings

60,000

700,000

(200,000 per tenant)

Number of VMACs

510

N/A

STP

All VLANs

N/A

Mis-Cabling Protocol (MCP)

2,000 VLANs per interface

12,000 logical ports (port x VLAN) per leaf

N/A

Mis-Cabling Protocol (MCP) (strict mode enabled on the port)

256 VLANs per interface

2,000 logical ports (port x VLAN) per leaf

N/A

Number of endpoints (EPs)

Default profile or High LPM profile:

  • MAC: 24,000

  • IPv4: 24,000

  • IPv6: 12,000

Maximum LPM profile:

  • MAC: 8,000

  • IPv4: 8,000

  • IPv6: 4,000

IPv4 scale profile:

  • MAC: 48,000

  • IPv4: 48,000

  • IPv6: Not supported

High Dual Stack scale profile:

  • LSE:

    • MAC: 64,000

    • IPv4: 64,000

    • IPv6: 24,000

  • LSE2:

    • MAC: 64,000

    • IPv4: 64,000

    • IPv6: 48,000

16-slot and 8-slot modular spine switches:

Max. 450,000 Proxy Database Entries in the fabric, which can be translated into any one of these:

  • 450,000 MAC-only EPs (each EP with one MAC only)

  • 225,000 IPv4 EPs (each EP with one MAC and one IPv4)

  • 150,000 dual-stack EPs (each EP with one MAC, one IPv4, and one IPv6)

The formula to calculate in mixed mode is:

#MAC + #IPv4 + #IPv6 <= 450,000

Note: Four fabric modules are required on all spines in the fabric to support above scale.

4-slot modular spine switches:

Max. 360,000 Proxy Database Entries in the fabric, which can be translated into any one of these:

  • 360,000 MAC-only EPs (each EP with one MAC only)

  • 180,000 IPv4 EPs (each EP with one MAC and one IPv4)

  • 120,000 dual-stack EPs (each EP with one MAC, one IPv4, and one IPv6)

The formula to calculate in mixed mode is:

#MAC + #IPv4 + #IPv6 <= 360,000

Note: Four fabric modules are required on all spines in the fabric to support above scale.

Number of endpoints (EPs)

(Continued)

High Policy profile:

  • LSE2 (except FXP switches):

    • MAC: 24,000

    • IPv4: 24,000

    • IPv6: 12,000

  • LSE:

    • MAC: 16,000

    • IPv4: 16,000

    • IPv6: 8,000

High IPv4 EP Scale profile:

  • LSE: Not supported

  • LSE2 (with 32GB of RAM):

    • MAC: 24,000

    • IPv4 local: 24,000

    • IPv4 total: 280,000

    • IPv6: 12,000

Multicast Heavy profile:

  • LSE: Not supported

  • LSE2 (except FXP switches):

    • MAC: 24,000

    • IPv4 local: 24,000

    • IPv4 total: 64,000

    • IPv6: 4,000

Fixed spine switches:

Max. 180,000 Proxy Database Entries in the fabric, which can be translated into any one of these:

  • 180,000 MAC-only EPs (each EP with one MAC only)

  • 90,000 IPv4 EPs (each EP with one MAC and one IPv4)

  • 60,000 dual-stack EPs (each EP with one MAC, one IPv4, and one IPv6)

The formula to calculate in mixed mode is:

#MAC + #IPv4 + #IPv6 <= 180,000

Number of Multicast Routes

Default (Dual Stack), IPv4 Scale, High LPM, High Policy or High IPv4 EP scale profiles: 8,000 with (S,G) scale not exceeding 4,000

Maximum LPM profile:

  • 1,000 with (S,G) scale not exceeding 500

High Dual Stack profile:

  • LSE: 512

  • LSE2: 32,000 with (S,G) scale not exceeding 16,000

Multicast Heavy profile:

  • LSE: not supported

  • LSE2 (with 32GB of RAM): 90,000 with (S,G) scale not exceeding 72,000

128,000

Number of Multicast Routes per VRF

Default (Dual Stack), IPv4 Scale, High LPM, High Policy or High IPv4 EP scale profiles: 8,000 with (S,G) scale not exceeding 4,000

Maximum LPM profile:

  • 1,000 with (S,G) scale not exceeding 500

High Dual Stack profile:

  • LSE: 512

  • LSE2: 32,000 with (S,G) scale not exceeding 16,000

Multicast Heavy profile:

  • LSE: not supported

  • LSE2 (except FXP switches): 32,000

32,000

IGMP snooping L2 multicast routes

  • For IGMPv2, route scale is for (*, G) only

  • For IGMPv3, route scale is for both (S, G) and (*, G)

Note

 

IGMP snooping entries are created per BD (2 receivers that join the same group from 2 different BDs consume 2 separate entries).

Default (Dual Stack), IPv4, High LPM, High Policy, or High IPv4 EP scale profiles: 8,000

Maximum LPM profile:

  • 1,000

High Dual Stack profile:

  • LSE: 512

  • LSE2: 32,000

Multicast Heavy profile:

  • LSE: not supported

  • LSE2: 32,000

32,000

Number of IPs per MAC

4,096

4,096

Number of Host-Based Routing Advertisements

30,000 host routes per border leaf

N/A

SPAN

32 unidirectional or 16 bidirectional sessions (fabric, access, or tenant)

N/A

Number of ports per SPAN session

Note

 

This is also the total number of unique ports (fabric and access) that can be used as SPAN sources across all SPAN sessions combined

63 – total number of unique ports (fabric + access) across all types of span sessions

N/A

Number of source EPGs in tenant SPAN sessions

Note

 

The numbers listed in this row assume that only tenant SPAN is configured.

If both, Access and Tenant SPAN are configured, this formula applies for both ingress and egress SPAN: 

E + P + E*P + EPP + v6FePP + 0.5*v4FePP <= 230

Where:

  • E— Number of source EPGs in Tenant SPAN

  • P—Number of source Ports in access SPAN without any filters

  • EPP—Number of (Epg,Port) Pairs in access SPAN with EPG filter only (no filter group)

  • v4FePP—Number of (v4 filter entry, Port) Pairs in access SPAN with filter group

  • v6FePP—Number of (v6 Filter entry, Port) Pairs in access SPAN with filter group

  • 230 bidirectional

  • 460 unidirectional (230 ingress + 230 egress)

N/A

Number of VLAN encapsulations per EPG

If EPG has 3 VLAN encapsulations = 3 entries

If EPG has 3 VLAN encapsulations = 3 entries

Number of SPAN ACL filter TCAM entries

SPAN filters are not supported in these:

  • Fabric ports

  • Fabric and tenant SPAN sessions

  • Spine switches

  • IPv4: 480

  • IPv6: 240

Total number of TCAM entries is calculated using this formula:

(IPv4-filters) * (IPv4-filter-source-groups) + 2 * (IPv6-filters) * (IPv6-filter-source-groups) + 2 * (no-filter-source-groups)

N/A

Number of L4 Port Ranges

16 (8 source and 8 destination )

First 16 port ranges consume a TCAM entry per range.

Each additional port range beyond the first 16 consumes a TCAM entry per port in the port range.

Filters with distinct source port range and destination port range count as 2 port ranges.

You cannot add more than 16 port ranges at once.

N/A

Common pervasive gateway

256 virtual IPs per Bridge Domain

N/A

Number of Data Plane policers at the interface level

  • 7 ingress policers

  • 3 egress policers

N/A

Number of Data Plane policers at EPG and interface level

128 ingress policers

N/A

Number of interfaces with Per-Protocol Per-Interface (PPPI) CoPP

63

N/A

Number of TCAM entries for Per-Protocol Per-Interface (PPPI) CoPP

256

One PPPI CoPP configuration may use more than one TCAM entry. The number of TCAM entries used for each configuration varies in each protocol and leaf platform. Use vsh_lc -c 'show system internal aclqos pppi copp tcam-usage' command to check on LSE/LSE2 platforms

N/A

Number of SNMP trap receivers

10

10

IP SLA probes*

*With 1 second probe time and 3 seconds of timeout

200

  • 1,500 (for PBR tracking)

  • 400 (for static route tracking)

First Hop Security (FHS)*

With any combination of BDs/EPGs/EPs within the supported limit

2,000 endpoints

1,000 bridge domains

N/A

Number of Q-in-Q tunnels

(both QinQ core and edge combined)

1,980

N/A

Number of TEP-to-TEP atomic counters

(tracked by 'dbgAcPathA' object)

N/A

1,600

SR-MPLS

Configurable Options

Per Leaf Scale

Per Fabric Scale

EVPN sessions

4

100

BGP labeled unicast (LU) pairs

16

200

ECMP paths

16

N/A

Infra SR-MPLS L3Outs*

* Including both, remote leaf and multi-pod

N/A

100 total, 2 per RL location

VRFs*

* Including remote leaf and multi-pod

800

5,000

External EPGs

800

5,000 total, 100 per VRF

Interfaces

N/A

Same as fabric scale

Multi-pod remote leaf pairs

N/A

50 pairs (100 RLs total)

Tenants

Configurable Options

Per Leaf Scale

Per Fabric Scale

Contexts (VRFs) per tenant

128 128

VRFs (Contexts)


Note

When deploying more than 1,000 VRFs, we recommend that all spines in the fabric have 32 GB of RAM.

All numbers are applicable to dual stack unless explicitly called out.

Configurable Options

Per Leaf Scale

Per Fabric Scale

Number of contexts (VRFs)

Default (Dual Stack) scale profile:

  • Switches with 32GB of RAM: 2,000

  • Other switches: 800

High Dual Stack, High LPM, High Policy scale profiles:

  • LSE2 switches with 32GB of RAM: 2,000

  • Other switches: 800

Maximum LPM scale profile:

  • LSE2 switches with 32GB of RAM: 250

  • Other switches: not supported

Multicast heavy, IPv4 and High IPv4 EP scale:

  • All switch models: 800

See Table 1: Fabric Scale Limits Per Cluster Size

Number of isolated EPGs

400

400

Border leaf switches per L3Out

N/A

24

Note

 

qualified with 100 VRFs + 16,000 IPv4 + 6400 IPv6 external prefixes

Number of vzAny provided contracts

Shared services: Not supported

Non-shared services: 70 per Context (VRF)

N/A

Number of vzAny consumed contracts

Shared services: 16 per Context (VRF)

Non-shared services: 70 per Context (VRF)

N/A

Number of graph instances per device cluster

N/A

500

L3Out per context (VRF)

N/A

400

Number of BFD neighbors

  • Up to 256 sessions using these minimum BFD timers:

    • minTx:50

    • minRx:50

    • multiplier:3

  • 257-2,000 sessions using these minimum BFD timers:

    • minTx:300

    • minRx:300

    • multiplier:3

N/A

Number of BGP neighbors

2,000 with up to 70,000 external prefixes with a single path

20,000

Number of OSPF neighbors

  • Up to 700 with up to 10,000 external prefixes using these timers:

    • Hello timer of 10 seconds

    • Dead timer of 40 seconds

    • No more than 300 OSPF neighbors per VRF

  • 701-2,000 with up to 35,000 external prefixe using these timers:

    • Hello timer of 40 seconds

    • Dead timer of 160 seconds

    • No more than 300 OSPF neighbors per VRF

12,000

Number of EIGRP neighbors

32

N/A

Number of subnets for route summarization

1,000

N/A

Number of static routes to a single SVI/VRF

5,000

N/A

Number of static routes on a single leaf switch

10,000

N/A

Number of IP Longest Prefix Matches (LPM) entries

Note

 

Except for the maximum LMP scale profile, the total of (# of IPv4 prefixes) + 2*(# of IPv6 prefixes) must not exceed the scale listed for IPv4 alone

Default (Dual Stack) profile:

  • IPv4: 20,000 or

  • IPv6: 10,000

  • IPv6 wide prefixes (>= /84): 1,000

    Note: This restriction only applies to EX models in LSE.

IPv4 scale profile:

  • IPv4: 38,000

  • IPv6: Not supported

High Dual Stack scale profile:

  • IPv4: 38,000 or

  • IPv6: 19,000

  • IPv6 wide prefixes (>= /84): 1,000

    Note: This restriction only applies to EX models in LSE.

 N/A

Number of IP Longest Prefix Matches (LPM) entries

(Continued)

Note

 

Except for the maximum LMP scale profile, the total of (# of IPv4 prefixes) + 2*(# of IPv6 prefixes) must not exceed the scale listed for IPv4 alone

High LPM Scale profile:

  • IPv4: 128,000 or

  • IPv6: 64,000

  • IPv6 wide prefixes (>= /84): 1,000

    Note: This restriction only applies to EX models in LSE.

Maximum LPM scale profile:

  • IPv4: 440,000

  • IPv6: 100,000

    Note: This profile also supports the combination of 440,000 IPv4 + 100,000 IPv6 prefixes.

High Policy profile:

  • LSE2 (except FXP switches):

    • IPv4: 20,000 or

    • IPv6: 10,000

  • LSE:

    • IPv4: 8,000

    • IPv6: 4,000

      Note: This restriction only applies to EX models in LSE.

High IPv4 EP Scale profile:

  • LSE2 (except FXP switches):

    • IPv4: 40,000

    • IPv6: 20,000

  • LSE: Not supported

Multicast Heavy profile:

  • LSE2 (except FXP switches):

    • IPv4: 20,000

    • IPv6: 10,000

  • LSE: Not supported

N/A

Number of Secondary addresses per logical interface

1

1

Number of L3 interfaces per Context

  • 1,000 SVIs

  • 48 Routed interfaces

  • 100 sub-interfaces with or without port-channel

N/A

Number of L3 interfaces

  • 1,000 SVIs

  • 48 Routed interfaces

  • 2,000 sub-interfaces with or without port-channel

N/A

Number of ARP entries for L3Outs

7,500

N/A

Shared L3Out

  • IPv4 Prefixes: 2,000 or

  • IPv6 Prefixes: 1,000

  • IPv4 Prefixes: 6,000 or

  • IPv6 Prefixes: 3,000

Number of L3Outs

2,000

See Table 1: Fabric Scale Limits Per Cluster Size

ECMP (Equal Cost MultiPath)

Configurable Options

Per Leaf Scale

Per Fabric Scale

Maximum ECMP for BGP

64

N/A

Maximum ECMP for OSPF

64

N/A

Maximum ECMP for Static Route

64

N/A

Number of ECMP groups

8,000

Note

 

Should not exceed 4,000 in steady state, to allow room for make-before-break transitions (*)

N/A

Number of ECMP members

Maximum LPM scale profile:

  • 64,000

Note

 

Should not exceed 32,000 in steady state, to allow room for make-before-break transitions (*)

All other scale profiles:

  • 32,000

Note

 

Should not exceed 16,000 in steady state, to allow room for make-before-break transitions (*)

N/A

Average number of paths (ECMP) per prefix at maximum LPM scale

Note

 

Across all prefixes, the average number of equal cost next-hops (ECMP) must not exceed the specified number. Some prefixes may have a higher number of paths as long as it's compensated by other prefixes that have a lower number of paths.

Default (Dual Stack), High Policy and Multicast Heavy profiles:

  • IPv4: 32

  • IPv6: 12

IPv4 scale profile:

  • IPv4: 16

  • IPv6: NA

High Dual Stack scale profile:

  • IPv4: 16

  • IPv6: 6

High LPM scale profile:

  • IPv4: 4

  • IPv6: 1

Maximum LPM scale profile:

  • IPv4: 1.8

  • IPv6: 1.8

N/A

Note

(*) For more information about managing the equal cost multipath scale, please see Understand and Manage ECMP Scale in Cisco ACI at this URL: https://www.cisco.com/c/en/us/solutions/collateral/data-center-virtualization/application-centric-infrastructure/manage-ecmp-scale-aci-wp.html.

External EPGs

Configurable Options

Per Leaf Scale

Per Fabric Scale

Number of External EPGs

  • Switches with 32GB of RAM: 2,000

  • Other switches: 800

See Table 1: Fabric Scale Limits Per Cluster Size

Number of External EPGs per L3Out

250

600

The listed scale is calculated as a product of (Number of external EPGs per L3Out)*(Number of border leaf switches for the L3Out)

For examples, 150 external EPGs on L3Out1 that is deployed on leaf1, leaf2, leaf3, and leaf4 adds up to a total of 600

Number of LPM Prefixes for External EPG Classification

Note

 

Maximum combined number of IPv4/IPv6 host and LPM prefixes for External EPG Classification must not exceed 64,000

Refer to LPM scale section

N/A

Number of host prefixes for External EPG Classification

Note

 

Maximum combined number of IPv4/IPv6 host and LPM prefixes for External EPG Classification must not exceed 64,000

Default Profile:

  • IPv4 (/32): 16,000

  • IPv6 (/128): 12,000

    Combined number of host prefixes and endpoints can't exceed 12,000

IPv4 Scale profile:

  • IPv4 (/32): 16,000

    Combined number of host prefixes, multicast groups, and endpoints can't exceed 56,000

  • IPv6 (/128): 0

N/A

Number of host prefixes for External EPG Classification

(Continued)

Note

 

Maximum combined number of IPv4/IPv6 host and LPM prefixes for External EPG Classification must not exceed 64,000

High Dual Stack Profile:

  • LSE:

    • IPv4 (/32): 64,000

      Combined number of host prefixes, multicast groups, and endpoints can't exceed 64,000

    • IPv6 (/128): 24,000

      Combined number of host prefixes and endpoints can't exceed 24,000

  • LSE2:

    • IPv4 (/32): 64,000

      Combined number of host prefixes, multicast groups, and endpoints can't exceed 64,000

    • IPv6 (/128): 48,000

      Combined number of host prefixes and endpoints can't exceed 48,000

High LPM Profile:

  • IPv4 (/32): 24,000

    Combined number of host prefixes, multicast groups, and endpoints can't exceed 24,000

  • IPv6 (/128): 12,000

    Combined number of host prefixes and endpoints can't exceed 12,000

N/A

Number of host prefixes for External EPG Classification

Note

 

Maximum combined number of IPv4/IPv6 host and LPM prefixes for External EPG Classification must not exceed 64,000

(Continued)

Maximum LPM profile:

  • IPv4 (/32): 10,000

    Combined number of host prefixes, multicast groups, and endpoints can't exceed 10,000

  • IPv6 (/128): 4,000

    Combined number of host prefixes and endpoints can't exceed 4,000

High Policy profile:

  • LSE:

    • IPv4 (/32): 16,000

      Combined number of host prefixes, multicast groups, and endpoints can't exceed 24,000

    • IPv6 (/128): 8,000

      Combined number of host prefixes and endpoints can't exceed 8,000

  • LSE2 (except FXP switches):

    • IPv4 (/32): 16,000

    • IPv6 (/128): 12,000

      Combined number of host prefixes and endpoints can't exceed 12,000

N/A

Number of host prefixes for External EPG Classification

Note

 

Maximum combined number of IPv4/IPv6 host and LPM prefixes for External EPG Classification must not exceed 64,000

(Continued)

High IPv4 EP Scale profile:

  • LSE: Not supported

  • LSE2 (except FXP switches):

    • IPv4 (/32): 16,000

    • IPv6 (/128): 12,000

      Combined number of host prefixes and endpoints can't exceed 12,000

Multicast Heavy profile:

  • LSE: Not supported

  • LSE2 (except FXP switches):

    • IPv4 (/32): 16,000

      Combined number of host prefixes, multicast groups, and endpoints can't exceed 154,000

    • IPv6 (/128): 4,000

      Combined number of host prefixes and endpoints can't exceed 4,000

N/A

Bridge Domains

Configurable Options

Per Leaf Scale

Per Fabric Scale

Number of BDs

1,980

Legacy mode: 3,500

15,000

Number of BDs with Unicast Routing per Context (VRF)

1,000

1,750

Number of subnets per BD

1,000, cannot be for all BDs.

1,000 per BD

Number of EPGs per BD

3,960

4,000

BD with Flood in Encapsulation: maximum number of replications (= EPG VLANs * ports)

The sum of all EPG VLANs * ports (i.e.,VLAN “replications”) for all EPG in a given BD with Flood in Encapsulation enabled must be less than 1,500

N/A

Number of L2 Outs per BD

1

1

Number of BDs with Custom MAC Address

1,000

1,000

Number of EPGs + L3Outs per Multicast Group

128

128

Number of BDs with L3 Multicast enabled

1,750

1,750

Number of VRFs with L3 Multicast enabled

64

300

Number of L3Outs per BD

16

N/A

Number of static routes behind pervasive BD (EP reachability)

N/A

450

DHCP relay addresses per BD across all labels

16

N/A

DHCP Relay: maximum number of replications (= EPG VLANs * ports)

The maximum number of VLAN encapsulations * ports in a BD with DHCP relay enabled should be less than 1,500

N/A

ICMPv6 ND: maximum number of replications (= EPG VLANs * ports)

The maximum number of VLAN encapsulations * ports in a BD should be less than 1,500

N/A

Number of external EPGs per L2 out

1

1

Number of PIM Neighbors

1,000

1,000

Number of PIM Neighbors per VRF

64

64

Number of L3Out physical interfaces with PIM enabled

32

N/A

Endpoint Groups (Under App Profiles)

Configurable Options

Per Leaf Scale

Per Fabric Scale

Number of EPGs

Normally 3,960; if legacy mode 3,500

15,000

Maximum amount of encapsulations per EPG

1 Static leaf binding, plus 10 Dynamic VMM

N/A

Maximum Path encap binding per EPG

Equals to number of ports on the leaf

N/A

EPGs with Flood in Encapsulation: maximum number of replications (= EPG VLANs * ports)

The sum of all EPG VLANs * ports (i.e.,VLAN “replications”) for all EPG with Flood in Encapsulation enabled in a given BD must be less than 1,500

N/A

Maximum amount of encapsulations per EPG per port with static binding

One (path or leaf binding)

N/A

Number of domains (physical, L2, L3)

100

N/A

Number of VMM domains

N/A

200 VDS

Number of native encapsulations

  • One per port, if a VLAN is used as a native VLAN.

  • Total number of ports, if there is a different native VLAN per port.

Applicable to each leaf independently

Number of 802.1p encapsulations

  • 1, if path binding then equals the number of ports.

  • If there is a different native VLAN per port, then it equals the number of ports.

Applicable to each leaf independently

Can encapsulation be tagged and untagged?

No

N/A

Number of Static endpoints per EPG

Maximum endpoints

N/A

Number of Subnets for inter-context access per tenant

4,000

N/A

Number of Taboo Contracts per EPG

2

N/A

IP-based EPG (bare metal)

4,000

N/A

MAC-based EPG (bare metal)

4,000

N/A

Contracts

Configurable Options

Per Leaf Scale

Per Fabric Scale

Security TCAM size

Default scale profile: 64,000

IPv4 scale profile: 64,000

High Dual Stack scale profile:

  • LSE: 8,000

  • LSE2: 128,000

High LPM scale profile:

  • LSE2 switches with 32GB of RAM: 32,000

  • Other switches: 8,000

Maximum LPM scale profile: 8,000

High Policy profile:

  • LSE: 100,000

  • LSE2 (with 24GB of RAM): 140,000

  • LSE2 (with 32GB of RAM): 256,000

High IPv4 EP Scale profile:

  • LSE: Not supported

  • LSE2 (except FXP switches): 64,000

Multicast Heavy profile:

  • LSE: Not supported

  • LSE2 (except FXP switches): 64,000

N/A

Software policy scale with Policy Table Compression enabled

(Number of actrlRule Managed Objects)

Dual stack profile: 80,000 (except EX switches)

High Dual Stack profile:

  • LSE: Not supported

  • LSE2: 140,000

High Policy profile:

  • LSE (except EX switches): 100,000

  • LSE2 (with 24GB of RAM) : 140,000

  • LSE2 (with 32GB of RAM) : 256,000

N/A

Approximate TCAM calculator given contracts and their use by EPGs

Number of entries in a contract X Number of Consumer EPGs X Number of Provider EPGs X 2

N/A

Number of consumers (or providers) of a contract that has more than 1 provider (or consumer)

100

100

Number of consumers (or providers) of a contract that has a single provider (or consumer)

1,000

1,000

Scale guideline for the number of Consumers and Providers for the same contract

N/A

Number of consumer EPGs * number of provider EPGs * number of filters in the contract <= 50,000

This scale limit is per contract.

If the limit is exceeded, the configuration is rejected.

If 90% of the limit is reached, fault returns.

Number of rules for consumer/provider relationships with in-band EPG

400

N/A

Number of rules for consumer/provider relationships with out-of-band EPG

400

N/A

Endpoint Security Groups (ESG)

Configurable Options

Scale

Number of ESGs per Fabric

10,000

Number of ESGs per VRF

4,000

Number of ESGs per Tenant

4,000

Number of L2 MAC Selectors per Leaf

5,000

Number of L3 IP Selectors per Leaf

5,000

Fiber Channel over Ethernet N-Port Virtualizer (FCoE NPV)

Configurable Options

Per Leaf Scale

Number of VSANs

32

Number of VFCs configured on physical ports and FEX ports

151

Number of VFCs on port-channel (PC), including SAN port-channel

7

Number of VFCs on virtual port-channel (vPC) interfaces, including FEX HIF vPC

151

Number of FDISC per port

255

Number of FDISC per leaf

1,000

Fiber Channel N-Port Virtualizer (FC NPV)

Configurable Options

Per Leaf Scale

Number of FC NP Uplink interfaces

48

Number of VSANs

32

Number of FDISC per port

255

Number of FDISC per leaf

1,000

Number of SAN port-channel, including VFC port-channel

7

Number of members in a SAN port-channel

16

VMM Scalability Limits

VMware

Configurable Options

Per Leaf Scale

Per Fabric Scale

Number of vCenters (VDS)

N/A

200 (Verified with a load of 10 events/minute for each vCenter)

Datacenters in a vCenter

N/A

15

Total Number of VMM domain (vCenter, Datacenter) instances.

N/A

200 VDS

Number of EPGs per vCenter/vDS

N/A

5,000

Number of EPGs to VMware domains/vDS

N/A

5,000

Number of endpoints per VDS

10,000

10,000

Number of endpoints per vCenter

10,000

10,000

Support RBAC for VDS

N/A

Yes

Number of Microsegment EPGs with vDS

400

N/A

Number of VM Attribute Tags per vCenter

N/A

vCenter version 6.0: 500

vCenter version 6.5: 1,000

Microsoft SCVMM

Configurable Options

Per Leaf Scale (On-Demand Mode)

Per Leaf Scale (Pre-Provision Mode)

Per Fabric Scale

Number of controllers per SCVMM domain

N/A

N/A

5

Number of SCVMM domains

N/A

N/A

25

EPGs per Microsoft VMM domain

N/A

N/A

3,000

EPGs per all Microsoft VMM domains

N/A

N/A

9,000

EP/VNICs per HyperV host

N/A

N/A

100

EP/VNICs per SCVMM

3,000

10,000

10,000

Number of Hyper-V hosts

64

N/A

N/A

Number of logical switch per host

N/A

N/A

1

Number of uplinks per logical switch

N/A

N/A

4

Microsoft micro-segmentation

1,000

Not Supported

N/A

Microsoft Windows Azure Pack

Configurable Options

Per Fabric Scale

Number of Windows Azure Pack subscriptions

1,000

Number of plans per Windows Azure Pack instance

150

Number of users per plan

200

Number of subscriptions per user

3

VM networks per Windows Azure Pack user

100

VM networks per Windows Azure Pack instance

3,000

Number of tenant shared services/providers

40

Number of consumers of shared services

40

Number of VIPs (Citrix)

50

Number of VIPs (F5)

50

Nutanix

Configurable Options

Per Fabric Scale
Total Number of Prism Central

10

Total Number of Nutanix domain instances

10

Number of EPGs per Prism Central

500

Number of EPGs per Nutanix domain

500

Number of endpoints per Prism Central (or Nutanix domain)

1,000

Number of VM Attribute Tags per Prism Central

500

Intra EPG isolation support per Prism Central

300 EPGs

Layer 4 to Layer 7 Services Scalability Limits

Configurable Options

Per Fabric Scale

Number of L4-L7 concrete devices

1,200

Number of graph instances

1,000

Number of device clusters per tenant

30

Number of interfaces per device cluster

Any

Number of graph instances per device cluster

500

Deployment scenario for ASA (transparent or routed)

Yes

Deployment scenario for Citrix - One arm with SNAT/etc.

Yes

Deployment scenario for F5 - One arm with SNAT/etc.

Yes

AD, TACACS, RBAC Scalability Limits

Configurable Options

Per Fabric Scale

Number of ACS/AD/LDAP authorization domains

4 tested (16 maximum /server type)

Number of login domains

15

Number of security domains/APIC

15

Number of security domains in which the tenant resides

4

Number of priorities

4 (16 per domain)

Number of shell profiles that can be returned.

4 (32 domains total)

Number of users

8,000 local / 8,000 remote

Number of simultaneous logins

500 connections / NGNIX simultaneous REST logins

Cisco Mini ACI Fabric Scalability Limits

Property

Maximum Scale

Number of spine switches

2

Number of leaf switches

4

Number of Pods

1

Number of tenants

25

Number of VRFs

25

Number of bridge domains (BDs)

1,000

Number of endpoint groups (EPGs)

1,000

Number of endpoints

20,000

Number of contracts

2,000

Number of service graph instances

20

Number of L4-L7 logical device clusters

3 Physical or 10 Virtual

Number of multicast groups

200

Number of BGP+OSPF sessions

25

GOLF VRF, Route Scale

N/A

Cisco ACI and UCSM Scalability

The following table shows verified scalability numbers for Cisco Unified Computing System with Cisco ACI ExternalSwitch app.

Configurable Options

Scale

Number of UCSMs per APIC cluster

12

Number of VMM Domains per UCSM

4

Number of VLANs + PVLAN per UCSM

4,000

Number of vNIC Templates per UCSM

16

QoS Scalability Limits

The following table shows QoS scale limits. The same numbers apply for topologies with or without remote leafs as well as with COS preservation and MPOD policy enabled.

QoS Mode

QoS Scale

Custom QoS Policy with DSCP

7

Custom QoS Policy with DSCP and Dot1P

7

Custom QoS Policy with Dot1P

38

Custom QoS Policy via a Contract

38

PTP Scalability Limits

The following table shows Precision Time Protocol (PTP) scale limits.

Configurable Options

Scale

(IEEE 1588 Default Profile)

Scale

(AES67, SMPTE-2059-2)

Scale

(Telecom Profile G.8275.1)

Number of leaf switches connected to a single spine with PTP globally enabled

288

40

N/A

Number of PTP peers per leaf switch

52

26

25

Number of ACI switches connected to the same tier-1 leaf switch (multi-tier topology) with PTP globally enabled

Within the range of the "Number of PTP peers per leaf switch" above

16

N/A

Number of access ports with PTP enabled on a leaf switch

Within the range of the "Number of PTP peers per leaf switch" above

Note

 

For improved performance on 1G interfaces with N9K-C93108TC-FX3P switches, the maximum number of 1G interfaces should not exceed 10

25

Note

 

For improved performance on 1G interfaces with N9K-C93108TC-FX3P switches, the maximum number of 1G interfaces should not exceed 10 out of 25

24

Number of PTP peers per access port

PTP Mode Multicast (Dynamic/Master): 2 peers

PTP Mode Unicast Master: 1 peer

PTP Mode Multicast (Dynamic/Master): 2 peers

PTP Mode Unicast Master: 1 peer

1

NetFlow Scale

Configurable Options

Per Leaf Scale

Number of exporters

2

Number of monitor policies under bridge domains

EX switches: 100

All other models: 350*

Number of monitor policies under L3Outs

EX switches: 100

All other models: 350*

Number of records per collect interval

EX switches: 20,000

All other models: 1,000,000**

* The total number of monitor policies under bridge domains and L3Outs must not exceed 350 (100 for EX switches).

** For more information, see Cisco APIC and NetFlow.