Hardware and Software Requirements for Branch

Bare-Metal Platforms

For bare-metal platforms, if hyperthreading is enabled in the BIOS, you must disable it. For information about how to disable hyperthreading, see the vendor manual for the white box.

To check whether hyperthreading is enabled, issue the following command:

admin@CPE:~$ lscpu | grep Thread

The output must show a value of 1. If the number of threads per core is 2 or more, hyperthreading is enabled. The following example output shows that hyperthreading is disabled:

admin@CPE:~$ lscpu | grep Thread
Thread(s) per core:    1

For platforms that serve a constant high rate of incoming traffic, such as hubs and large branch devices, it is recommended that you issue the following command to isolate the CPU, which optimizes the platform's performance. After you issue this command, you must reboot the device.

cli> request system isolate-cpu enable

For more information about the hardware requirements for bare-metal platforms, contact Versa Networks Customer Support.

Virtual Machine Platforms

For virtual machine (VM) deployments, you must allocate a dedicated CPU and memory (with 1:1 provisioning) to the Versa branch components using 100% CPU and 100% memory reservation. It is recommended that you do not overprovision the hypervisor, meaning that the sum of virtual resources for VMs should not exceed the physical resources. Depending on the number of sockets present in the host, you can create a non-uniform memory access (NUMA) node with affinity for the VM.

To improve deterministic performance, it is recommended that you set CPU Latency Sensitivity to High to prevent the scheduler from performing dynamic vCPU-to-pCPU reassignment. It is recommended that you turn off hyperthreading at the host level, by disabling hyperthreading in the host's BIOS. When hyperthreading is enabled, the number of available cores doubles. To verify the number of active cores, issue the lscpu command in the host's operating system.

If you are using VMWare, please refer to the related VMWare recommendations.

For NIC interface allocation, it is recommended that you use PCI passthrough or single-root input/output virtualization (SR-IOV) to achieve maximum performance. If it is not possible to use PCI passthrough or SR-IOV, use VMXNET3 on VMWare, and VirtIO on KVM/OpenStack. Note that when the VMXNET3 or VirtIO performance limit is reached, you can add additional VMXNET3 or VirtIO interfaces to horizontally scale the VNF network performance.

The VM's CPU, memory and NIC allocation for the VNF should be allocated from the same NUMA nodes to achieve predictable performance and latency. If the VM performs the Versa Controller control plane function, allocating the secondary NIC from a different NUMA node is acceptable for NIC resiliency. This should have a minimal impact on control plane performance.

Note that you do not need to perform any of the above steps on public cloud platforms such as AWS, Azure, or GCP, as long as you use the instances recommended by Versa Networks. The instances recommended by Versa Networks on public cloud platforms already come with all of the above configurations.

The following table lists the VM software supported for VOS devices.