VMware on Monday released the first public spin of its ESXi bare-metal hypervisor for selected 64-bit Arm systems.
Dubbed the ESXi-Arm Fling, this non-production-grade, unsupported technical preview can be downloaded here.
It's primed and ready for the Raspberry Pi 4 Model B (the 8GB variant is highly recommended) as well as Ampere eMAG-powered Avantek workstations and Lenovo rack-mounted servers, the NXP LX2160A-based HoneyComb LX2K, and the LS1046A-based NXP Freeway. You can find the software's requirements and fine-print here. A My VMware account is needed to fetch the installation files.
Once up and running on your hardware, you can boot up guest Arm operating systems, such as Ubuntu Linux 20.04, and run applications on top of them as you'd expect from a hypervisor. It can be managed as a standalone host or from a vCenter Server 7.0 or later instance. If you're experienced with VMware environments, it should feel and act like any other host system, plus or minus any missing functionality.
The aim of the ESXi-Arm Fling is to help developers and enterprise architects evaluate the technology for all sorts of applications, from processing at the network edge to SmartNICs to mobile app build-and-test servers, and see how it could slot into an existing VMware environment, if need be.
Imagine something like a multi-core Raspberry Pi or HoneyComb LX2K in the field running a guest operating system that communicates with the outside world via an internet link and via a touchscreen user interface, and another guest environment simultaneously running a real-time OS that takes inputs from sensors, makes decisions, and outputs control signals to other equipment. Seeing as it's virtualized, you have a bit more flexibility in updating or restarting the UI guest without interrupting that main guest doing all the important sensor and control work, and also you can update, shutdown, or spin up a redundant copy of the main guest from the UI environment.
VMware and Nvidia buddy up to integrate AI containers with vSphere, promising results as 'fast as public cloud'READ MORE
Then there's VMware's SmartNIC dream – code-named Project Monterey – in which PCIe cards are loaded with one or more accelerators – think GPUs and FPGAs – and controlled by a cluster of bolted-on Arm CPU cores. These cards can be plugged straight into a network, and offload workloads such as storage management, security monitoring, connection routing, and packet filtering from the host's microprocessors to the card's acceleration engines. This involves ESXi running on those Arm cores to virtualize tasks, carve up resources, and integrate with VMware Cloud Foundation.
"As the name suggests, ESXi-Arm is a version of ESXi built to run on 64-bit Arm processors," said Kit Colbert, VMware's cloud platform business unit CTO, who added that Virtzilla had worked with Arm to get its port working.
"Traditionally, ESXi was designed for x86 compute virtualization. However, a few years ago, a small team of Arm experts at VMware started working on porting ESXi to Arm. This was a huge technical undertaking. Not only is ESXi a large codebase, with many technical subtleties, but the underlying semantics of Arm processors are quite different than x86. This required careful testing and modification to ensure ESXi ran on both Arm and x86 platforms while maintaining ESXi’s high standards."
With the Raspberry Pi 4 with 8GB now available, we have an easily-accessible hardware platform for ESXi-Arm
Colbert went on: "We have wanted to release an ESXi-Arm Fling for quite some time. However, the ESXi code wasn’t quite ready and there wasn’t a lightweight, widely-proliferated Arm platform that fully met our requirements. We’re happy to say that both of these issues have been addressed! We’ve matured the ESXi-Arm code such that it works outside of the highly controlled environments we leveraged for demos and implemented broader device driver and platform support. And with the Raspberry Pi 4 with 8GB now available, we have an easily-accessible hardware platform for ESXi-Arm."
To make sure everyone gets the message this is pre-release software, the ESXi-Arm Fling will expire 180 days after installation, though it may be reinstalled again. For what it's worth, a fling in this context is a tool or app built by VMware's engineers, or a community effort, intended to be used as an experiment or early evaluation. A bit of fun, perhaps.
The Register reported the official emergence of the ARM64 port of the hypervisor during last week's VMworld event, and earlier predicted its arrival. The ESXi-Arm Fling was announced on the opening day of Arm's DevSummit, a virtual event running this week. We assume Nvidia approves.
And, it almost goes without saying, ESXi-Arm isn't the only Arm hypervisor out there. Xen just announced an official port of its hypervisor to the Raspberry Pi 4, Linux KVM on Arm is a thing, and Wind River's VxWorks can manage it, for instance. ®