Next year, VMware plans to launch a rearchitected vSphere that can offload security, AI-related data processing and network functions from a server CPU to specialized hardware. The new form of server virtualization would boost infrastructure efficiency significantly, company officials said.
VMware unveiled the latest vSphere technology this week at its VMworld 2020 virtual conference. The re-architected virtualization infrastructure would offload some application services to specialized chips running on what IDC calls “function accelerator cards.” The transfer would save as much as 20% of a CPU’s capacity, VMware said.
Regardless of size or industry, all VMware customers would likely have a use for the new architecture, dubbed Project Monterey, IDC analyst Ashish Nadkarni said. “Whoever is a VMware shop today would benefit from it.”
VMware CEO Pat Gelsinger described Project Monterey to reporters as a “pretty major re-architecture of vSphere.” VMware partners Intel, Nvidia and Pensando would make the chips that would power the separate services. Dell Technologies, Hewlett Packard Enterprise and Lenovo would develop the supporting server hardware.
VMware referred to Project Monterey devices as smart network interface cards (smart NICs). They would power VMware’s ESXi hypervisor, which would provide the platform for running the offloaded services.
But Nadkarni said smart NICs’ old definition no longer applies to the hardware VMware plans to use. “Smart NICs is an old term, and it does not do justice to the work that Pensando and the likes are doing,” he said.
Project Monterey like AWS Nitro
VMware’s architecture is similar to AWS’ Nitro system, which offloads networking, storage and management overhead from the CPU to a dedicated hypervisor, Nadkarni said. VMware’s Project Monterey is an example of converting hyperscale technology into a mainstream enterprise product.
For VMware customers, Project Monterey will let them run more applications on a single server, thereby reducing computing costs, said Paul Turner, senior director of product management at vSphere. Project Monterey also pools the application resources running on specialized hardware. That enables IT managers to use VMware’s vCenter to manage them across software running on CPUs.
Another advantage of using a network of processors separate from CPUs is a more reliable infrastructure, Nardkani said. If an application server goes down, it won’t affect services running on the specialized hardware, which could be in a separate device or on the same motherboard as the CPU.
“That almost creates a [separate] mini-brain in the server,” Nardkani said.
Exactly what application services VMware customers can run on Project Monterey next year isn’t entirely clear. Firewalls that inspect packet payloads are on tap. Still, other offerings will depend on what can run on the specialized system-on-a-chip (SoC) delivered by VMware partners.
Gelsinger said he expects enterprises and telcos to use the technology. The latter will likely run network functions for enterprise services delivered over 5G wireless technology.
The SoCs will consist of components, including GPUs and data processing units (DPUs), a technology developed by network device maker Mellanox Technologies. Nvidia acquired Mellanox this year for $6.9 billion.
“All of those together are part of this next wave of an entirely distributed GPU, CPU, DPU set of resources,” Gelsinger said. “The vSphere layer will be able to manage and optimize [all of them].”