Network Configuration for Nutanix Cloud Platform with FlashArray

Nutanix

Audience
Public
Product
FlashArray
FlashArray > Purity//FA
Technology Integrations
Source Type
Documentation
Customers can deploy the infrastructure and configure Nutanix NCI compute nodes, Everpure FlashArray, and the intervening network in multiple ways. Examples include dedicated switches for each VLAN, switches carrying multiple VLANs over trunk links, or switches configured in vPC/MLAG mode. You should configure the Ethernet interfaces on the Nutanix nodes in Active/Active with MAC pinning for NVMe/TCP traffic. Nutanix and Everpure do not recommend configuring Ethernet interfaces using No-Bond modes, but as with many recommendations there can be exceptions. If your environment may require this configuration please reach out to your Nutanix team to confirm. In some environments, multiple layers of switches or routers may exist between the compute nodes and the storage arrays.
Note: For Cisco based deployments, please refer to the Cisco Validated Design for more information.

Layer2 VLAN Segmentation

The recommend networking configuration from both Nutanix and Everpure is to use at least two VLANs dedicated to NVMe/TCP data storage traffic to achieve traffic isolation and fault tolerance. This approach provides the following benefits:

  • Resiliency: A failure or maintenance of a single VLAN does not impact all I/O paths.

  • Operational Isolation: Broadcast storms, excessive traffic, or Layer 2 loops are contained within a single VLAN domain.

  • Simplified Troubleshooting: VLAN-level isolation enables faster fault isolation when diagnosing connectivity or performance issues.

  • Balanced Load Distribution: Traffic can be more evenly spread across network uplinks and switches when VLANs map to different physical paths.

Although NVMe/TCP traffic can be distributed across more than two VLANs, the practical benefits of doing so are limited. Beyond separating traffic into additional Layer 2 broadcast domains, this approach does not provide meaningful improvements in performance, resiliency, or fault isolation for NVMe/TCP workloads.

The examples in the later section each VLAN uses a /24 subnet, supporting up to 254 usable IP addresses. Depending on deployment scale and design requirements, a larger subnet may be appropriate.

Layer 2 Spanning-Tree Protocol Configuration

We recommend that you keep the Spanning Tree Protocol (STP) enabled for all Layer 2 broadcast domains. To achieve faster convergence during link flaps or switch reloads, interfaces connected to the nodes or the storage arrays should be configured as edge ports. This allows the ports to transition immediately to the forwarding state without passing through the STP blocking or learning states upon linkup.

The following text shows a sample configuration for a Cisco Nexus 9000 switch:
Nexus9k(config)# interface ethernet 1/1
Nexus9k(config-if)# spanning-tree port type edge

We strongly discourage disabling STP because doing so introduces significantly more risk than benefit in an NVMe/TCP-based storage environment. STP provides essential loop protection, and its absence can lead to broadcast storms, traffic black-holing, and widespread I/O disruption.

VLAN Tagging

There are a different options available to customers when configuring Nutanix or FlashArray NVMe/TCP networking. There are some benefits to enabling VLAN tagging on switch ports by configuring them as trunk ports that allow only the required VLANs to the compute nodes and storage arrays. That is, you should configure VLAN tagging between the nodes and the switches and between the switches and the arrays. See the network diagram of trunk links with subnetting the section FlashArray Configuration for Integration with Nutanix Cloud Platform of this guide.

This approach ensures consistent, end-to-end VLAN tagging across the Layer 2 domain, while restricting trunks to the necessary VLANs enhances security and stability.

For more information, see Deploying NCI Compute Cluster with Everpure FlashArray on the Nutanix website for more information.

IP Addressing

We recommend the following best practices to ensure stable connectivity and predictable behavior for NVMe/TCP deployments between Everpure FlashArray and Nutanix nodes. Adhering to these guidelines helps avoid IP-related disruptions and ensures proper Layer 3 network services in redundant switch designs.

  • To prevent IP address changes after a reboot—for example, when using dynamic assignment through DHCP—we recommend configuring static IP addresses for each NVMe/TCP interface on both the Everpure FlashArray and the Nutanix nodes.

  • If you must use DHCP, ensure that DHCP reservations are configured so that IP addresses remain consistent across reboots.

  • When switches are configured in vPC/MLAG mode, configure Switched Virtual Interfaces (SVIs) for the NVMe/TCP VLANs as required to provide Layer 3 (network) services.

Switch/Network Configuration Recommendations

To achieve optimal performance and predictable behavior for NVMe/TCP traffic, you should design the underlying network with low latency, sufficient bandwidth, and consistent configuration across all components. The following recommendations help ensure efficient packet forwarding and proper traffic prioritization.
  • Use cut-through switching mode (rather than store-and-forward) to minimize network latency.

  • Leverage a non-blocking switching architecture to avoid congestion under sustained I/O loads.

  • Ensure MTU consistency across hosts, switches, and storage arrays: for example, 9216 bytes within the network and 9000 bytes at the endpoints.

  • Configure QoS/DSCP markings to prioritize storage traffic using recommended values (DSCP26 or AF31).

  • Do not enable Priority Flow Control (PFC) or Global Pause, as these mechanisms are not supported by Everpure FlashArray in NVMe/TCP environments.