System requirements

Before installing BIG-IP Next for Kubernetes, make sure your environment meets the requirements in this section.

Important

  • Only one DPU per chassis is supported for BIG-IP Next for Kubernetes. Extra DPUs in the chassis are reserved for GPU-to-GPU or GPU-to-storage acceleration.

  • The DPU must be dedicated to BIG-IP Next for Kubernetes. Do not install other software on the DPU, as this can cause performance issues or Kubernetes node instability.

Note

F5 does not validate the accuracy or safety of any third-party links listed below.

Supported DPU platforms

The following DPU platform is supported.

Note

See the NVIDIA DOCA documentation for instructions on checking and updating the firmware version.

NIC

Driver version

Min. firmware version

NVIDIA BlueField‑3 (model B3220, both Crypto Enabled and Crypto Disabled)

DOCA 3.2.0 (host and DPU)

32.47.1026

Hardware requirements

Category

Component

Requirement

Notes

DPU

DPU

(See Supported DPU platforms.)

A k8s node. Must be dedicated to BNK. Do not run other software on the DPU.

Memory

DPU Memory

Minimum 16 GB

TMM requires hugepages.

Memory

Host Memory

Minimum 32 GB

For application workloads.

Storage

Storage

NFS Server

Persistent volumes required for dSSM, OTEL, and core file storage.

Network

Network Setup

Dual NIC (internal and external)

Requires two OVS bridges on the DPU: internal and external.

Software requirements

Software

Version

Notes

DOCA

3.2.0

Install on the DPU and server. Includes drivers, firmware, and runtime.

RShim

Included with DOCA

Enable RShim on the server to access the DPU management console.

Helm

v3.x

Used to install FLO and FLP.

Calico CNI

v3.27.0

Primary CNI. Other CNIs may work but are not tested.

Multus

Latest

Required to attach SR‑IOV interfaces to TMM pods.

SR‑IOV Network Device Plugin

Latest

Exposes DPU scalable functions to the Kubernetes scheduler.

NFS CSI Driver

v4.10.0

Storage provisioner for persistent volumes.

cert‑manager

Latest

Manages TLS certificates for Zero Trust mTLS between components.

Minimum Gateway API version

When installing a BNK (DPU) instance on bare metal or a VM using F5 Lifecycle Operator, you’ll specify in the CNEInstance manifest that you want to later configure the BNK instance using Gateway API (and the BNK Gateway API supplemental CRDs).

BNK requires Gateway API v1.4.1 experimental or later.

If the Gateway API CRDs

Then F5 Lifecycle Operator

Are not installed in the cluster

Will install an appropriate version

Are already present in the cluster

Will not alter those CRDs

Note

You must ensure the Gateway API CRDs are maintained at a version appropriate for BNK.

Network planning

Plan the following network resources before installation.

Resource

Requirements

Example or notes

External VLAN IPs

One Self IP for each TMM instance. Supports IPv4 and IPv6.

Example: 11.19.1.80–82 for three TMMs

Internal VLAN IPs

One Self IP for each TMM instance. Supports IPv4 and IPv6.

Example: 10.19.1.80–82 for three TMMs

SNAT Pool IPs

IPs used for egress SNAT.

One pool per egress setup

Virtual Server IPs

IPs for ingress virtual servers or F5 IPAM‑assigned.

Can be auto-assigned by the F5 IPAM Controller

OVS Bridge Config

Two bridges per DPU: internal and external.

Set up with ovs-vsctl

Scalable Functions

Minimum two SFs per DPU (one internal, one external).

F5 recommends one or more SFs per PF

IPv6 Static Config

Set IPv6 on the host and DPU before installing Kubernetes.

Required if k8s is dual‑stack cluster

MTU Plan

MTU consistent across VLANs, up to 9000.

Set in the F5SPKVlan custom resource

Capacity planning and sizing

BNK performance depends on DPU count, TMM resource allocation, and workload needs.

Component

Default

Notes

TMM Pods per DPU

1 (DaemonSet)

One TMM pod on each DPU in Whole Cluster mode.

TMM CPU

Guaranteed QoS class

CPU requests match limits.

TMM Memory

Hugepages required

Set hugepages on the DPU. TMM uses DPDK for packet processing.

dSSM Replicas

3

High availability through Redis Sentinel.

OTEL Collector

1 replica

Prometheus exporter listens on port 9090.

CWC

1 replica

Manages licensing, telemetry, and debug APIs.

CNE Controller

1 replica

Watches all namespaces in Whole Cluster mode.

Note

For throughput and connections‑per‑second sizing, contact your F5 account team.

Hugepages

The amount of hugepage memory to configure on the DPU depends on the CNE’s deploymentSize. See CNEInstance CR. The deployment size sets the number of CPUs and threads the f5-tmm pod uses. F5 recommends 2 mebibyte (MiB) hugepages (hugepages-2Mi). Note that 1 mebibyte (1 MiB) equals 1,048,576 bytes, which is not the same as 1 megabyte (1 MB). This table shows the hugepages-2Mi page count for each deployment size.

CNE-instance deployment size

Hugepages-2Mi page count

Small

2048

Medium

4096

Large

8192

Max

12300

Next steps