Skip to content

Fabrics

Fabrics are isolated, self-contained arrangements of network interconnections that form a topology, including the switches, links, and logical structure.

The Fabrics window gives administrators an overview of each Fabric by displaying the objects in a collapsible tree view.

Fabrics are organized into tree-node containers called Fabric Collections.

Colors🔗

  • The default color for a disabled fabric object is grey.
  • The default color for an enabled working fabric object is green.

Alarm Colors🔗

When an object triggers an alarm, its icon changes to the alarm's color . The color change cascades up the hierarchy, always reflecting the most severe active alarm at each level. In the image below, the Fabric named Compute Fabric has child device(s) triggering a severe (red) alarm. This notification cascades through the entire hierarchy to the Fabric's parent.

How to Create a Fabric Collection🔗

  1. Select Fabric Collections.
  2. Click Add Fabric Collection.
  3. In the form field that appears provide a name and click Add Fabric Collection.

Creating Fabrics🔗

  1. Select the parent Fabric Collection.
  2. Click the Add Fabric button. ()
  3. A form appears requiring a name for the Fabric and a Fabric type. Fill in the fields and click Add Fabric.

The hierarchical object components display differently depending on the fabric type. The tree contains nodes that represent fabrics, categories for switch devices (LAGs, Pods, SuperSpines, Leafs, etc.), and the switch devices themselves.

Fabric Types🔗

Management

A switch network providing connectivity to the out-of-band management ports of switches across the other fabrics.

  • Fabrics
    • Management-Example
      • Switches

Data Center 3 Stage Clos

Spine-Leaf network configuration where traffic traverses three switch hops — leaf → spine → leaf — within a single POD.

  • Fabrics
    • Data-Center-3-Example
      • Lags
      • PODs
      • Super Spines

Data Center 5 Stage Clos

Spine-Leaf network configuration traversing five hops — leaf → spine → superspine → spine → leaf — across multiple PODs. Superspines interconnect PODs for non-blocking any-to-any connectivity.

  • Fabrics
    • Data-Center-3-Example
      • Lags
      • PODs
      • Superspines

Multiplane

  • Fabrics
    • Multiplane-Example
      • Lags
      • PODs
      • Spine Planes

Packet Broker

A Packet Broker is a network device placed between network traffic sources and monitoring tools (or other destinations).

  • Fabrics
    • Packet-Broker-Example
      • Egress
      • Ingress

Spectrum-X Fabric Types🔗

Editing Spectrum-X Capacity Settings🔗

For Spectrum-X fabrics, the Max SUs and SU Size fields define the scale and capacity used for switch unit, leaf, spine, and HGX connection planning. These fields are editable only while the Fabric is empty.

After the Fabric has user-added content, Verity disables these fields to protect existing topology and connection calculations. Hover over the disabled field to view the reason, such as Cannot change Max SUs when the Fabric has content or Cannot change SU size when the Fabric has content.

To change these values, first empty the Fabric of user-added content. Auto-created empty planes and, for Spectrum-X 2 Tier fabrics, the single auto-created empty tier-2 POD do not prevent the Fabric from being treated as empty.

Spectrum-X 2 Tier and Spectrum-X 3 Tier fabrics can be deployed as single-plane or multi-plane systems. Each plane is an independent, complete network fabric. In multi-plane configurations, each SuperNIC's single high-speed port is split into multiple lower-speed links — one per plane — so the GPU sees full bandwidth while traffic is distributed across redundant, isolated fabrics for increased scale and resilience. In Topology planes are displayed as rectangular boxes .

Spectrum X Cabling Validation🔗

To verify that a Spectrum X system's physical topology is cabled correctly, complete the following steps:

  • From Reports, select Missing Port Connections.

Note

If this Missing Port Connections Report is empty, it means the system did not find anything wrong with the cabling configuration.

The information is presented as three columns. These are color coded as white, pink and peach. These columns have containing rows and each represents a physical connection point listed as a Source Port. The device and switch point associated with the Source Port are also listed as Source Device and Source Switchpoint. The first two columns (White and Peach) list pairs of devices that are disconnected from one another. A technician is expected to physically cable these devices according to the connection points presented. The third column (titled Excluded Connections) lists incorrect connections, meaning links that exist but are not part of the intended topology.A technician is expected to physically disconnect these connections.

Example

In the image below, port _rail0 on device EW2-hgx-su00-h000 should connect to port 1s0 on device EW2-leaf-su00-r0. That connection is missing. Instead, _rail0 is incorrectly connected to port 1s0 on a different device, EW2-leaf-su00-r1.

Spectrum-X 2 Tier

NVIDIA Spectrum-X Ethernet networking platform. Can be configured as a multi-plane system for increased redundancy and scale.

  • Fabrics
    • Spectrum-X-2-Tier-Example
      • Spines
      • SUs

Spectrum-X 3 Tier

NVIDIA Spectrum-X Ethernet networking platform. Can be configured as a multi-plane system for increased redundancy and scale.

  • Fabrics
    • Data-Center-3-Example
      • PODs
      • Superspine Groups

Inspector Window Object Details🔗

The inspector window to the right reveals information and parameters of the selected tree object.

Fabric Details🔗

Fabric objects have customizable fields that appear when you click them.

Managed Pairs🔗

The Managed Pairs button () opens a window that lets you manage and create Switch Pairs. Switch Pairs can be created as Multi Chassis redundancy pairs, or as non-redundant L3 leaf switch pairs when Multi Chassis is not selected.

Verity allows the pair to use leaf endpoints across PODs.

General🔗

Fabric Type and other settings including DHCP Snooping.

Reserved Ranges🔗

Provisioning reserved ranges specific to the fabric are listed here. Other Reserved Ranges are available in Admin → System → Reserved Ranges

Feature Flags🔗

This section lists additional configuration options including settings to allow underlay connections between Pods, port admin and port status interval values.

Underlay Configuration🔗

Manages fundamental network infrastructure settings, including a shared Anycast MAC address for redundancy and load balancing. It controls how network devices handle address aging and connection restoration, ensuring stable and efficient baseline network communication.

Leaf Underlay Fabric Configuration🔗

These settings control how leaf devices establish and maintain BGP connections, including timing for connection checks, routing updates, and connection establishment, ensuring stable and responsive network edge communications.

Spine Underlay Fabric Configuration🔗

For data center CLOS fabrics, spine switches in the same fabric may use the same BGP AS number. Verity allows duplicate BGP AS numbers between spine endpoints in the same CLOS fabric, including when the fabric-level Spine AS Number is not set. Superspine endpoints in the same fabric may also share the same BGP AS number.

Defines how spine devices maintain and establish BGP (Border Gateway Protocol) connections. These settings manage network behaviors such as connection keepalive intervals, connection timeout periods, and routing information exchange.

Data Timer Configuration🔗

Provides advanced timing controls that optimize connection stability and performance. These settings allow network administrators to fine-tune how devices handle interface state changes, link detection, and MAC address management. By adjusting Link State Timeout, Startup Delay, and MAC Holdtime, users can customize network resilience and reduce unnecessary connection disruptions.

Route Aggregation🔗

Combines multiple specific IP routes into a single summarized route advertisement, so neighboring routers only need to know one prefix instead of many.

QoS🔗

QoS is a set of techniques that prioritize certain types of network traffic over others when bandwidth is constrained or congestion occurs.

DSCP to TC Map🔗

DSCP to TC Map is a translation table that converts quality-of-service markings between two different systems. This feature maps incoming DSCP priority markings (0-63) to internal Traffic Class queues (0-7) to ensure high-priority traffic like voice and video receives proper handling.

Click the Priority numbers below the DSCP numbers (image needed) to choose the range of numbers to set (image needed).

Notes🔗

A persistent note writing section for administrators.

Container Details🔗

Some tree view items exist only to group their children. Clicking these nodes reveals aggregate details about the child objects. The following image shows that the node named External contains 3 LAGs.

LAGs (Fabric/External)🔗

You view and manage LAG parameters from the Fabrics navigation window.

Fabric LAG Parameters🔗

External LAG Parameters🔗

Switches (Spine,Leafs)🔗

Status🔗

These indicators provide information about the device state.

  • Comm Status: Indicates the overall communication health and connectivity state of the device within the management system.

  • Provisioning Status: Indicates the current provisioning state.

  • Device Management Status: When you enable Device Management parameters, this indicates the connection status.

  • State in SDLC: Indicates SDLC behavior.

Location and Type🔗

Type of device (Spine, Leaf, Superspine, Management) , POD and Rack information is listed here.

Network Settings🔗

Network Settings vary depending on the device type.

Switch Info🔗

Switch Info displays device hardware details, such as make and model, and basic configuration information, including the current management IP address. This information is also viewable from Topology at the top of Switch ().

Device Management🔗

Device Management settings are editable and fall under Switchpoint settings. In addition to the parameters in Device Management, Switchpoint settings include all port provisioning of the switch. You can copy these values between devices using the Copy Provisioning feature.

How to Create a Switch Pair From Fabrics🔗

  • Determine the target devices. If the target devices do not yet exist, create them before creating a switch pair.
  • Select the chosen Fabric or POD to apply the Switch Pair.
  • In the upper portion of the page, click Manage Pairs.

  • Enter a name in the Name field, then select your devices under the Switchpoint fields.
  • Use the Multi Chassis checkbox to control whether the pair is a redundancy pair. The checkbox is selected by default for data center fabrics. Clear the checkbox to create a non-redundant L3 leaf switch pair.

When complete, the Fabrics tree displays the devices as a related pair with the Switch Pair icon. Peer LAG, MCLAG, and pair link status handling apply only to pairs where Multi Chassis is selected.

How to Create a Fabric Gateway🔗

A Fabric Gateway represents each end of a planned BGP session between fabrics. For high availability, four fabric gateways are needed — one for each end of each session.

  1. From Fabrics, select the Fabric.
  2. Select Gateways.
  3. In the upper right corner, click Add Gateway.

Configure each Gateway object with the following:

  • The Neighbor AS number of the remote Leaf switch
  • The IP address assigned to the Gateway Port Profile on that neighbor's switch
  • The Fabric Interconnect option enabled

Gateway Modes🔗🔗

Gateways can be one of several different modes. The Gateway Mode field defines how the software handles and routes traffic within the network. It determines whether routing decisions are based on a statically defined default gateway, static routes, traditional BGP routing, or new enhancements to BGP to support dynamic neighbors.

  • Default: A single static default route.
  • Static: Uses manually configured static routes, no dynamic routing.
  • Static BGP: Establishes BGP sessions with manually configured BGP neighbors. The BGP neighbors are explicitly defined by their IP addresses, and no dynamic peer discovery occurs within the subnet.
  • Dynamic BGP: BGP neighbors are dynamically discovered within a subnet, leading to the creation of multiple BGP sessions from a single tenant gateway object.