darren whittingham - Fotolia
A new partnership between VMware and Nvidia brings a more efficient way to live migrate vGPU-based workloads. This capability is a part of vSphere version 6.7, update 1, which will be available in October 2018.
Being able to live migrate workloads that use virtual graphics processing units (vGPUs) is a slick way for companies to make the most out of their GPU hardware investments. This new VMware partnership announced at VMworld 2018 gives Nvidia live migration support for GPU-accelerated VMs on both Citrix XenMotion -- which has been available since early 2018 -- and VMware vMotion.
The ability to move workloads around and meet spikes in processing capacity demands is a primary reason customers use virtualization. For example, a billing reconciliation application may require a tremendous amount of after-hour CPU cycles. With live migration, admins can move this workload to a high-performing server when it needs the additional cycles and move it off when the workload is idle, letting other processes take advantage of the increased computing power.
Expanded GPU-based workflow support
GPUs reduce processing time from months and weeks to hours or even minutes. Common applications include machine learning, ray tracing and graphics processing. This decreased calculation time is why some businesses are willing to pay for a GPU-based infrastructure, as well as why more hardware manufacturers sell GPUs and a reason VMware supports the hardware.
But most businesses can't afford to go buy racks of GPUs and dedicate one per workload; this is where the live migration feature of the VMware and Nvidia partnership adds value. With the ability to host and live migrate vGPU-based VMs, companies can buy fewer GPU cards and still get accelerated processing power.
Admins can now use vMotion to live migrate Nvidia vGPU-enabled virtual desktops to another server without any data loss or end-user interruptions.
Furthermore, if a server runs too many virtual desktops and starts to become overloaded, admins can live migrate the desktop to another GPU-accelerated server for instant performance improvements via one mouse click.
Along with leveling workloads, VMware live migration is useful for hardware and software upgrades without VM downtime. IT professionals can work at a time that's convenient for them, instead of having to come into the office during the early morning hours when no one is in the office.
This means depending on the time of day, admins can switch workloads off a vGPU-based infrastructure and start a different workload that needs the processing capabilities. A good example of this is a data analysis or simulation project that takes several weeks to run. Admins can dedicate GPU-accelerated servers to virtual desktops during the day when employees are at work, and then migrate the data analysis workload to the server with GPUs after hours. This allows the task to complete on the highest-performing infrastructure without affecting end users.
Support VDI with VMware live migration
The ability to live migrate workloads using vGPUs is useful for virtual desktop infrastructure (VDI) workloads. Many customers hold off upgrading their existing VDI deployments or shifting to VDI because of how resource intensive it is. With live migration capabilities, businesses can expand and cost effectively upgrade VDI deployments, which might drive higher adoption of GPU-accelerated VDI.
Virtual GPU live migration also brings infrastructure resiliency because it''s possible to extend the capabilities between data centers over a layer two network. This means if a primary data center experiences failure, the workloads, including the end users' VDI sessions, are live migrated across the wide area network to another data center without any interruptions.
Virtual GPU live migration benefits are not just limited to mixed artificial intelligence, machine learning and VDI workloads. IT admins also benefit from being able to keep servers healthy, as well as implement patches and updates on their own time without disruption to users, and can run AI and ML processes on different servers.
This partnership between VMware and Nvidia should drive greater vGPU adoption at many organizations that want this feature.
Note: VMware and Nvidia are clients of ZK Research.