Last year at this time, VMware had just announced Project Pacific, an initiative to build container-orchestration platform Kubernetes into VMware’s virtualization centerpiece, vSphere. This week at a virtual VMworld 2020, the company delivered on that goal and charted a course to the next generation of enterprise computing.
VMware has married its core server virtualization platform and cloud infrastructure platform with its Tanzu portfolio, VMware’s branded platform for cloud-native app development that comprises Kubernetes, an Istio-based service mesh, Mission Control and other related services. Together, they give admins and users a single interface to deploy virtual machines along with container-based applications.
The company has been talking about its vision for hybrid and multi-cloud for a long time, but it took Kubernetes to make it a reality. In his opening remarks, CEO Pat Gelsinger recalled an earlier foundational technology that paved the way for modern computing.
"Much like Java two decades ago, Kubernetes is a rare technology that brings everyone together," he said.
Kubernetes At Your Service
Though all the pieces fit together like "fish and chips," Gelsinger said, VMware officials acknowledge that these are still early days for vSphere with Tanzu and Cloud Foundation 4 with Tanzu.
The first two of four editions – Basic and Standard – of the Tanzu portfolio integration are now available. They are the first steps in VMware’s attempt to make Kubernetes more "consumable" and deliver a better Kubernetes "product experience," said Craig McLuckie, vice president of research and development at VMware, in a video interview. "The Basic edition is intended for our admins who are increasingly being asked to deliver a Kubernetes infrastructure environment, so this provides a very simple baseline expression of that."
It is not known when the Advanced and Enterprise Tanzu editions will be available, but when they do come out, they are going to be focused on power users – large companies or service providers who want a more DevOps-oriented and automated experience for armies of developers.
The VMware Cloud stack is now being hosted by all the major cloud providers. This week, VMware also announced Tanzu portfolio support for VMware Cloud on AWS, and preview support for Oracle Cloud VMware Solution. Officials said VMware is working with Microsoft on a version for the Azure cloud.
VMware User Group (VMUG) President Steve Athanas said he thinks VMware is doing a good job walking a fine line with customers – innovating with leading-edge technology and offering new functionality in its Tanzu portfolio while continuing to support the legacy operations. "They definitely are ahead of where the customers are now [in cloud-native], and they're doing a good job leading them along, holding their hand," he said in a video interview. "This is if you want to build new applications. This is the way to do it. [VMware is saying] We're not making you do it. But it's easy to do it if you want to."
Project Monterey
What kinds of apps will be developed is a still-evolving question, but the company talked a lot about a new project that promises to open opportunities for running enterprise apps and development. The way VMware officials put it, Project Monterey is a response to the demands emerging applications are putting on traditional server infrastructures.
Project Monterey, at its core, is about the porting of the ESXi hypervisor to the ARM chip platform, which was a multi-year project. This has now enabled VMware to run its virtualization platform on ARM chips, specifically those found on emerging SmartNICs – network interface cards with additional CPUs, DPUs and GPUs.
VMware has partnered with NVIDIA, which acquired network card maker Mellanox earlier this year. Mellanox’s Bluefield DPU (data processing unit) will be the among the first products to be able to host VMware ESXi instances running vSphere, NSX network virtualization, vSAN storage, security and management services. VMware has also partnered with Intel and P4-chip maker Pensando, as well as server vendors Dell, HPE and Lenovo. Products are expected to come out next year.
Server Disaggregation
The Project Monterey technology will be like having a second server on the server, which VMware officials say will be able to improve performance of the core x86 server CPU by offloading cryptographic, data, or AI algorithm processing to the NIC. Also, the SmartNICs can function on their own network, isolated from the central server. vSphere would enable admins to manage the x86 and ARM workloads and applications.
"[With Project Monterey,] we will be able to migrate all the network components or the storage components and the host management component from the hypervisor that was running in the x86 into this SmartNIC," said Kit Colbert, CTO of VMware’s Cloud Platform, in a video interview. "What that provides is a much more resilient and self-contained system that can provide capabilities to the system regardless of what's going on in the application side."
The NICs running VMware software will appear in general purpose servers, likely for larger enterprises or telecom firms, and in hyperconverged systems for small and midsize enterprises. Over the long term, officials acknowledge that the SmartNICs will enable new use cases, including pushing some compute functions out to the network "edge" via container applications or serverless functions.
"Because of the work with Project Pacific and the integration of Kubernetes with vSphere, each ESXi host has a kubelet [Kubernetes node] already on it," Colbert said. "Kubernetes is a very strategic focus for us across the board, up and down our stack, and I think it'll be the same here as well."
VMUG’s Athanas, who doubles as Associate CIO for System Architecture at the University of Massachusetts-Lowell, thinks Project Monterey is "gee-whiz" technology at this point but sees the potential, especially for offloading security management.
"In the server space, one of the things that we're struggling with is needing to manage the security software and agents and policies on every machine individually," said Athanas. "That takes up a lot of CPU cycles that aren't going to production applications."
