At VMware, we simplify how customers deploy and run workloads, run on the best available infrastructure, run efficiently, all wrapped around a familiar management experience. Coming back from Arm Techcon 2019, we saw a tremendous amount of innovation on display for running workloads. And in collaboration with Arm and some of its partners, we are also exploring several use cases for running our ESXi hypervisor on various Arm SoCs, at the Edge, on the SmartNIC, and in the Cloud, with the same set of principles to help our customers efficiently run and manage their workloads.
(Arm NEOVERSE investments over the next several years)
ESXi on Arm at the Edge
At the Edge, where the environment is often difficult to reach and compute resources are limited, it’s important to have infrastructure that can provide resiliency in a compact form factor. These environments are not operated by the central IT teams, but rather by field operating teams whose servicing cost is dependent on number of truck-rolls than by the cost of the hardware. To explore this use case, we showcased running our hypervisor on a compact Marvell Armada A8040-based SolidRun MacchiatoBin providing vSphere vMotion and Fault Tolerance inside a wind turbine scenario. When there’s a need to patch the hardware, or even if one of the hardware nodes goes down, our technology enables the workload to continue running uninterrupted. We bring the same data center-grade resiliency to the Edge.
The second characteristic of Edge environment is that the compute resources are limited, either due to the need to bring down the cost per node or limited power availability. Our ESXi hypervisor is known for deploying on the big iron running monster workloads. But the same efficient hypervisor is just as good scaling down to small machines running small workloads. We explored just how small by running it on the Raspberry Pi4. Our goal is to bring data center virtualization benefits such as vMotion, Fault Tolerance and lifecycle management into environments where these capabilities are not readily available.
ESXi on Arm on the SmartNIC
Bringing the discussion back a bit closer to home base, we’re also exploring helping our customers more efficiently run their data center by leveraging hardware accelerators. Some of our work in this area include virtualization support for GPUs and FPGAs, as well as remote acceleration through Bitfusion. An emerging accelerator is the SmartNIC, where the NIC is now equipped with a capable general-purpose processor and plenty of ram. This enables customers to offload data plane processing such as running the virtual switch, security or storage functionality, or even analytics on the incoming traffic, all without interrupting the CPU. The goal here is to offload tasks to optimized hardware and focus the CPU to run the customer’s actual workload. In environments where network processing can chew up to 20-30% of the CPU cycles, SmartNICs can help customers can achieve higher consolidation ratios AND faster workload execution. You can have your cake and eat it too.
(Showcasing ESXi running on various Arm SoCs, including Mellanox BlueField, Marvell Armada, Raspberry Pi4)
ESXi on Arm in the Cloud
If we dial back time a bit, all the way to November of 2018, AWS announced the availability of the Arm-powered EC2 A1 VM instance and more recently the availability of the A1 metal instance. The biggest draw for customers is the price point, up to 40% more cost effective than comparable x86 instances. The workloads targeted include webservers such as NGINX which can take advantage of the cores to scale-out, containerized services, filers which are often waiting for I/O and not compute heavy, and applications written in interpreted languages. In addition to the cost benefit, developers also gain access to flexible compute consumption and the availability of the AWS ecosystem.
For VMware customers leveraging VMware Cloud on AWS (VMC), this new A1 metal instance could potentially complement their existing x86 instances. For example, customers can run cost-optimized workloads such as NGINX on a cluster of ESXi-powered A1 hosts, working alongside or communicating with a cluster of ESXi-powered x86 hosts running traditional database applications. The key here is that customers can choose the platform(s) best suited for their application mix, managing this heterogeneous environment with operational consistency via the vSphere control plane. We’re pretty excited with this use case and actively exploring it with AWS and Arm:
“The announcement of the Arm Neoverse-based EC2 A1 bare metal instances provides new opportunities for customers to leverage VMware ESXi in the Cloud,” said Mohamed Awad, vice president of Marketing, Infrastructure Line of Business, Arm. “Our collaboration with VMware in this customer evaluation phase is a step forward in enabling solutions for the next generation of Cloud-to-Edge workloads.”
(ESXi running on the AWS a1.metal instance managed by vCenter Server. The same vCenter Server is also managing various other ESXi Arm hosts and can also include ESXi x86 hosts.)
We’re still early days in the above Cloud effort, as well as continuing to explore new ways we can bring the benefits of our infrastructure stack to customers. The ESXi on Arm bits are not currently available to the public. However, if the use cases above are of interest in your organization, we’d like to hear your feedback and explore POC opportunities, please reach out to your account team to get in touch.