Nutanix AOS 5.20 LTS released supporting vGPU live migrations
On 16th of May, Nutanix released their new AOS 5.20 LTS software. This AOS LTS release comes also packaged with a new version of the Nutanix AHV hypervisor: AHV-20201105.2030. This new LTS version release is very important to many customers as it enables “Acropolis Dynamic Scheduling (ADS) for VM’s with vGPU’s”. Yes, that means live migrations of VM’s with vGPU’s is now also available for AOS LTS users!
What’s new in Nutanix AOS 5.20
ADS support for VM’s with vGPU’s
The new version of AHV (AHV-20201105.2030) supports Acropolis Dynamic Scheduling (ADS) for VM’s with vGPU’s. When a number of VM’s with vGPU’s are running on a host and you enable ADS support for the cluster, the Lazan manager invokes VM migration tasks to resolve resource hotspots or fragmentation in the cluster to power on incoming vGPU VM’s. The Lazan manager can migrate vGPU-enabled VM’s to other hosts in the cluster subject to certain conditions.
Console Support for VM with only one vGPU
Like other VM’s, you can access VM’s with vGPU’s also using the console. Enable or disable console support for a VM with only one vGPU configured. Enabling console support for a VM with multiple vGPU’s is not supported. By default, console support for a VM with vGPU’s is disabled.
NVIDIA Host Driver v12.0 Support
Nutanix AOS 5.20 also supports the latest NVIDIA Host drivers: v12.0. This was also long awaited as customers were still forced to use the older v11 or even v10 drivers on their previous AOS STS or LTS software versions. Both of those driver versions are now EOL according to NVIDIA.
Information on the drivers can be found on the website from NVIDIA available at: https://docs.nvidia.com/grid/12.0/index.html.
Do note that downloading the NVIDIA Host drivers requires to be done using the Nutanix Support & Insights Portal: https://portal.nutanix.com/page/downloads?product=ahv. On that page, select “NVIDIA” from the dropdown list on top allowing you to see and download the available v12.0 drivers.
vGPU Live Migrations on Nutanix is not new
The ability to use ADS in a Nutanix cluster is not something new. This feature was already made available with the releases of AOS 5.18.1 & 5.19 combined with the v11.1, v11.2 and v11.3 NVIDIA Host drivers. But those AOS versions are STS releases, which many customers are not using on their Nutanix clusters. These customers have chosen AOS LTS software release because they want an AOS version, which is maintained for a longer duration with primarily bug fixes. Having the latest and greatest features accompanied with the need to update more frequently (STS) is not their “cup of tea”.
More information on LTS vs. STS is available at: https://portal.nutanix.com/kb/000005505.
Upgrading Nutanix AOS, AHV & NVIDIA Host Drivers
Upgrading from an earlier version of Nutanix AOS and AHV requires to be performed via the Life Cycle Manager (LCM) feature within Nutanix Prism.
Do ensure that you first have updated LCM but also Foundation and NCC before proceeding with the other upgrade tasks.
Nutanix recommends that you upgrade to the latest supported AHV version as soon as possible after having upgraded AOS. As mentioned before, AHV is bundled with AOS and therefore available for upgrade without the need for an extra download. After having upgraded AOS, your cluster can temporarily run the unsupported or older AHV version until you complete the AHV upgrade. See the Nutanix Compatibility Matrix for your reference: https://portal.nutanix.com/page/documents/compatibility-matrix.
However, before actually upgrading AHV you will need to first uninstall the existing NVIDIA Host Drivers. Please, see the following Nutanix KB article explaining this process: https://portal.nutanix.com/kb/7973.
When the old NVIDIA Host Drivers have been uninstalled, you can proceed with the AHV upgrade and finally the installation of the new NVIDIA Host Drivers v12.0
Conclusion
Nutanix AOS 5.20 LTS has been a long awaited software release. Previously when having to perform maintenance on Nutanix AHV VDI host it was required to shutdown hosted vGPU VM’s because of the unavailability of ADS. That meant downtime and the requirement to perform the activity outside normal working hours. Finally, AOS LTS users can also benefit from performing vGPU VM’s live migrations!
PS. It goes without saying (but I will state it anyway) that you will need sufficient host resources in your Nutanix cluster to offload vGPU VM’s from one host to others in the same cluster. 😉