VMware has defended its decision to end public downloads of its Virtual Disk Development Kit (VDDK). VDDK is an SDK that a Broadcom technical document describes as “a set of libraries and utilities that enable software developers to create applications that can access and manipulate virtual disks used” and also “provides APIs for backup and recovery, cloning, and other operations on virtual disks.” Those qualities mean VDDK is also useful when moving VMs from VMware to another virtualization environment. VMware’s rivals spotted that and made VDDK part of their migration tools. Nutanix’s Move, for example, requires a VDDK library to do the job. Red Hat also recommends using VDDK for migrations. ShapeBlue, an integrator that specializes in Apache CloudStack, an alternative to VMware, rates VDDK as an important part of migration tools. “The virt-v2v engine that sits underneath most VMware-to-KVM conversion work supports VDDK as an input transport,” the company wrote in late August. ShapeBlue’s post appeared a day or two after VMware, without fanfare or an explanation, took down the web pages that offered VDDK downloads. “A vendor is entitled to change how it distributes an SDK,” ShapeBlue wrote. “Doing it without a word, while the Apache CloudStack administration guide and other projects’ install instructions still tell people to click a link that no longer resolves, is something else.” “The cost lands unevenly too,” the company added. “A large backup vendor with a partner relationship has someone to call. A three-person migration consultancy, an open-source project, or a customer converting their own workloads has a forum post and a 404.” In an email to The Register, VMware defended its actions. To understand its response, know that VMware calls third party developers who work with its products technology alliance partners – aka “TAPs”. “Access to the VDDK is still available through select TAP for its licensed use case, which has always been backup and recovery,” VMware told The Register. “Customers needing backup and recovery capabilities can choose from the many approved solutions already available from these partners.” “The VDDK is not a customer entitlement. It is a software development kit covered by its own developer license and was never included with a Broadcom software purchase,” a company spokesperson added. VMware’s argument is therefore that users should not expect they can use VDDK, and that migrations is an unlicensed use case for the tool – which remains available as part of products from TAP members. VMware’s rivals have pointed out that ending public VDDK downloads could disrupt some migration projects that have already commenced, making the withdrawal of downloads jarring given the virtualization giant hasn’t previously policed use of the SDK. Services organizations that perform VMware migrations have found workarounds that make it possible to migrate VMs without needing VDDK. VMware’s decision to stop public VDDK downloads comes at a time when its changed licensing policies have seen some users either consider reducing their use of its products – as was the case at VMware channel partner Rackspace – or shifting to a rival platform. VDDK has been downloadable for years and VMware cannot have been ignorant of how its rivals used the SDK to facilitate migrations. It’s therefore not hard to conclude that this is a defensive move by VMware, given it allowed downloads for years. Moving VMs is, however, among the easier parts of a virtualization migration project. Rebuilding or replacing the entirety of a virtualized environment, across networks, security, and storage, is the hard part and VMware knows it - which is why it offers a site recovery tool that doubles as a migration utility. Removing easy access to VDDK therefore doesn’t change much for those planning a move away from VMware. ®

The price of eggs and butter might have soared in the last couple of years, but if you want a real sticker shock horror story, ask a VMware customer. After Broadcom acquired VMware in 2023, some of them complained about hypervisor licenses rising by up to 600 percent. Add to that the difficulty in refreshing hardware thanks to supply-chain delays, and the enterprise AI plan your board is asking for begins to look like a distant dream. Migrating comes with costs too. It isn’t just a compute-level move; networking, storage, management tools, and security all come along for the ride. Gartner found projects taking 18 to 48 months, at $300 to $3,000 per VM. On October 7 at The Brewery in London, Nutanix will help you tackle the migration problem and set up a solid AI strategy at the same time. .NEXT on Tour London is a free, day-long event for infrastructure leaders, cloud architects, platform engineers and datacenter managers. It’s a chance to share coffee with your peers and lament the pain of being a tech leader during one of the craziest moments in the industry, while exploring the way forward. Ideally, you’ll walk away with a migration blueprint at the end of the day. The details of everyone’s blueprint will vary, but one thing will likely be the same: a software-first approach. With new tin currently costing two limbs and your first-born, Nutanix will counsel you to sweat your assets or harness the cloud. Deploying software that ties together and manages your existing infrastructure will be a key theme, with sessions on external storage integrations and hardware reuse. It doesn’t matter if you’re doing it on-premises or in the cloud, or both; the executive keynote from Nutanix chief technology officer Manosiz Bhattacharyya focuses on choice. Making migration hurt less is only one goal for the day. The other is working out an AI strategy that you can pursue with the money you’ll save. Everyone’s talking about this but it’s a challenge; only 37% of AI initiatives over the past 12 months were classified as live and delivering value, per S&P Global. You can learn how to control token costs and tackle the scourge of shadow AI with Nutanix Agentic Gateway. If migrating your infrastructure and readying it for AI seems daunting, take a breath and listen to the guest speaker Bonita Norris, the youngest person to reach both the North Pole and the summit of Everest. That’ll give you some perspective. Attendance is free. Secure your place here. Sponsored by Nutanix.

EXCLUSIVE What's old is new again as VMware will soon release an updated version of vSphere Standard, the low-end server virtualization bundle that it hasn’t significantly changed for years and has scarcely promoted since its 2023 acquisition by Broadcom. VMware’s hero product for the last three years has been the Cloud Foundation (VCF) private cloud bundle. vSphere Standard and another low-end suite called vSphere Enterprise Plus remained on VMware’s list of products, but the website mentioning the products devotes a handful of words to each. The Register has often heard VMware, and its partners, would not issue subscription renewal quotes for the low-end products or only offer quotes that suggested adopting VCF instead. VCF is more powerful than standalone vSphere but more complex and costly – and also overkill for basic server virtualization. We’ve also heard that Broadcom sold another smaller bundle – vSphere Foundation – almost exclusively to customers that also acquired VCF but needed something smaller than VCF for some sites. Speaking to The Register at the VMware Explore conference on Wednesday, Paul Turner, chief product officer for VMware’s Cloud Foundation Division, said that the Broadcom business unit has changed the incentives that saw salespeople steer customers toward VCF. “We corrected this,” Turner said, adding that VMware has sometimes had “too big a focus on VCF.” VMware’s last major release of vSphere Standard came in 2022 with version 8. In 2025, VMware delivered version 9 of vSphere along with VCF 9 – but didn’t release a new cut of vSphere Standard. Turner said that decision was taken because VMware chose to focus on improving the security of VCF. Ram Velaga, president of Broadcom’s Infrastructure Software Group, told The Register VMware decided to focus on VCF because it wanted to shift the prevailing narrative that public clouds are the natural home for workloads, and instead argue that private clouds are more cost effective and easier to operate. Bringing a new low-end server virtualization offering to market at that time may have confused customers, he told The Register. “People could get distracted,” he said, suggesting that VMware could have prompted questions about the extent of its commitment to private clouds. Velaga said vSphere Standard is suited to users who operate around 128 cores. Turner mentioned memory tiering as a possible feature of the product. As it happened, Turner and Velaga’s remarks came on the same day that Proxmox, the provider of an open-source virtualization and containerization platform, announced it had opened a North American office and started to offer 24x7 support for the first time. The Austria-based company previously only supported its wares during local business hours. The Register often hears vSphere users mention Proxmox as an ideal replacement for low-end server virtualization. Turner said the new vSphere Standard will be a better and more resilient offering than Proxmox and pointed to the introduction of 24x7 support as a sign of Proxmox’s maturity being well behind that of VMware and its partners. He also said that more details about the new vSphere Standard will likely emerge as VMware takes its Explore conference to Germany, in mid-October. Among the facts he said will emerge soon are how VMware will bring vSphere Standard to market, an item of interest as the Broadcom business unit dropped the majority of its channel partners. Earlier this week, The Register predicted VMware would not make a pitch to its many low-end users at the Explore conference. The imminent release of a vSphere Standard upgrade was not made on stage at the event, but during one-to-one interviews – so perhaps we were technically correct! Turner thinks another of our assertions, that VMware was not interested in lower-end users, was incorrect. We leave it to readers to make their own judgment about the level of interest in small customers VMware displayed by spending three years focusing on VCF. Low-end VMware users will likely be relieved and intrigued by news of a vSphere Standard revival. Few VMware users wanted to quit the product, which has a deserved reputation for working brilliantly. Yet many felt the need to acquire VCF, and the cost of that package, meant it was necessary to consider VMware alternatives. VMware’s competitors saw that thinking as an opportunity: HPE, for example, even made its low-end virtualization bundle free for a year. Such offers will soon be less potent, because an upgrade is always easier than a migration. ®

Visiting a second bar on the same night can often be a dangerous move. When Dave Bradley did it, he ended up having to move a datacenter, on a two-week deadline, without extra budget. Bradley is a senior cloud engineer at healthcare services company Vizient and visited a bar while winding down after an off-site company meeting. After enjoying a decent stint of merriment with colleagues from the company’s tech team, Bradley decided to call it a night - but on his way out, he spotted his boss in a second watering hole, conversing with other even more senior execs. He decided to stop in and stayed a while before the second group decided their night was over. As Bradley walked out, one of the senior suits asked, “Could we shut down the Chicago datacenter in two weeks and get everything moved down to Dallas?” Bradley’s reaction was “What?” Then his boss intervened and asked if it was possible. Helped by some top shelf liquor, Bradley and a colleague from the networking team started to brainstorm the project. Management didn’t hold Bradley to the two-week timeframe, but settled on a demanding six-to-seven-month target. Speaking at the VMware Explore conference on Monday, Bradley said Vizient used Pure Storage (now Everpure) arrays and had 10Gbps links between the two datacenters. Replicating data between arrays was the obvious move, and the company had sufficient bandwidth to make it happen. Of course it wasn’t that easy. The two datacenters had different IP addressing schemes, and the company didn’t have enough hardware to support all workloads during the downtime that a physical hardware migration would cause. Magic tricks In one meeting, Bradley says execs told him “Just make it work. You guys are good. So we got to be magicians.” In another meeting, Bradley decided to own the name. Asked for details of the project, he replied, “Just trust the magicians.” He’s now known around the office as “The magician.” One trick he and the team used to solve the hardware issue was consolidating workloads onto a subset of the servers and storage arrays that Vizient ran in its Chicago datacenter. That process freed up some hardware, which the company shipped to Dallas so they would be in place to receive workloads during the migration. Bradley came to Vizient after a merger, and in his previous role had built a full disaster recovery rig on VMware Site Recovery Manager (SRM), Virtzilla’s tool that makes it possible to shift workloads between two sites that both run the vStack. Vizient still had licenses to use it, so he put it to work because it can remap IP addresses to the local scheme during migrations. That those licenses were paid for was very helpful, because the migration budget was minimal. More big nights By now, Bradley’s team worked alongside the networking team and others across the company’s IT. Together they planned a three-night migration push, each devoted to a different part of the company’s infrastructure. The first night saw them shift Vizient’s development environment, which comprised about 70 VMs and three VLANs. Bradley said he and his team created orchestrations that saw workloads shut down in Chicago, before new VLANs spun up in Dallas ready to run the apps once they landed. “We built in a couple check spots where it would pause so we could verify things were right,” he said. “Our network guys would move the VLAN from Chicago, bring it up down in Dallas, and we would click the button to bring it back up.” It worked and Bradley said he was pleased, but shocked. Management then threw a spanner in the works by acquiring a company and deciding its workloads also needed to move to Dallas. Thankfully, that company was a tenant of the same Chicago office building and datacenter as Vizient, had only 50 VMs, and was in the midst of a project that meant they were ready to move. Bradley still had unused hardware on hand and literally moved an array into the acquired company’s offices, copied relevant data, brought it back to Vizient and added it to the migration plan. During this phase of the operation Bradley learned that turning off a Pure array involves pulling out its power cord, a method he had to confirm by checking with support! Night two also went well and saw the company’s test environment move without issue. It's always DNS Next came the attempt to move the 50 VMs from the recently acquired company, during which Bradley said: “Bad stuff happens.” The aphorism “It’s always DNS” proved accurate and the migration team learned that SRM can glitch when asked to move a server that hosts both DNS and Active Directory, and the DNS goes down. “So in the middle of it, we had to figure out what was going on,” Bradley said. The company’s network team figured out that they would need to leave some workloads in Chicago until a later date. Bradley described the final push to move the company’s production environment as “The most anticlimactic end to a project I ever had.” The job took three-and-a-half hours and all plans worked as intended. The final move resulted in about two hours of downtime, and the others produced about an hour each. “Most of the business never realized it happened because it went so well,” Bradley said. He thinks the project went so well because his approach to tech work complemented those of colleagues – some are detail-oriented, others more likely to just attack a problem. Together they found a balanced approach to the problems the project presented. Bradley offered three takeaways from his experience. One was that transport can be a bottleneck: courier company Federal Express would not issue insurance to allow a single shipment of hardware between Chicago and Dallas, because the value of the equipment exceeded its policy limits. Vizient therefore needed to make multiple shipments of storage arrays and servers. Happily, the company’s datacenter is close to a Chicago airport, so arranging multiple shipments was not complicated. The second learning was that DNS is always a problem. “Sometimes, projects happen after three or four cocktails,” was the third note of caution he mentioned in his talk. Maybe that’s something to remember if you ever think about going into a second bar! ®

Arm servers won’t have significant market share in the enterprise for at least three to five years, according to Ram Velaga, the recently installed leader of Broadcom’s software business. That’s unwelcome news for hardware vendors contemplating Arm servers, because buyers will be less interested in the devices if they can’t manage them with the tools they already use. VMware remains the most commonly deployed enterprise server virtualization supplier, so its users won’t be able to embrace Arm servers anytime soon without Arm boxes becoming a silo – a scenario few will embrace. Velaga previously led Broadcom’s networking silicon group and became president of Broadcom’s enterprise software division on January 1, putting him in charge of the former VMware, Symantec, and CA business units. Speaking at VMware’s Explore conference in Las Vegas on Tuesday, Velaga said the virtualization pioneer’s mission won’t change on his watch. “We will be the abstraction layer, for CPU and GPU, regardless of instruction set,” he said. VMware currently virtualizes servers, storage, and networks, but only on x86 servers. The Broadcom business unit has gone a little wider with GPUs, supporting Nvidia, AMD, and Intel. Another device VMware works with is the SmartNIC/Data Processing Unit – network cards that pack a processor so they can handle networking chores to save the CPU the trouble. The processor in many SmartNICs uses the Arm architecture, so VMware created a version of its hypervisor to run in SmartNICs as part of its distributed services engine, which runs workloads like firewalls and intrusion detection tools on the devices. While VMware worked on that project, it also inched toward creating a hypervisor for Arm servers, and recently delivered a preview of that tech. In his remarks during his conference-opening plenary on Monday, Velaga suggested Arm is likely to become part of the enterprise IT scene. But during a Q&A session on Tuesday he shared his view that it will be a few years before Arm servers have significant market share in the enterprise, and that VMware therefore doesn’t need to port its hypervisor in a hurry. Velaga added his view that VMware customers have zero interest in processors built on RISC-V, the permissively licensed instruction set that some pundits believe could do to processor design what Linux did to proprietary operating systems. VMware’s mission to abstract everything means it will address RISC-V if it makes sense to do so in future, but for now the business unit isn’t touching it. Velaga also outlined how Broadcom software spends its research and development budget and said around 60 percent of its $5 billion annual spend goes towards keeping VMware’s current products up to date. The remainder goes toward developing new products such as the "AI factory" announced yesterday. AI is hot. Arm servers are not. It’s therefore not hard to see where VMware will spend its R&D dollars in coming years. ®

VMware is working to find new roles for its memory tiering technology, in the expectation that memory and hardware prices will remain high for years and users will therefore look for ways to put off server purchases. Memory tiering moves data out of DRAM and onto another storage medium that’s fast enough that CPUs can still access info at speeds that don’t slow application performance. VMware’s implementation works by looking for “cold pages” in DRAM – chunks of memory containing data that a CPU hasn’t touched and/or won’t need for a while. When a CPU can’t find any capacity in DRAM, it can pause operations until memory becomes available. That leaves CPUs and expensive DRAM idling wastefully when they could be churning through work. To stop the waste, memory tiering looks for cold pages in DRAM and moves them to another storage medium – for VMware that means fast and resilient NVMe drives. The virtualization pioneer introduced memory tiering last year in version 9 of Cloud Foundation, its bundle of products that allow users to assemble and operate a private cloud. In this year’s VCF 9.1 update, VMware tweaked it to improve performance when running databases – always a key workload – and added dashboards that show users the extent of tiering activity, plus more management tools. VMware delivered memory tiering just as the AI boom saw the price of DRAM and solid state storage skyrocket. DRAM now often costs several times more than the servers it lives in. At the VMware Explore conference on Monday, staff technical marketing architect Dave Morera said VMware plans to support more VM types with the tech. “We have a two-to-three year roadmap,” he said in a conference session. Some of the items on that roadmap will emerge in the VCF 9.2 release that VMware will likely debut around May 2027. Morera said VMware’s memory tiering currently supports about 75 percent of workloads and wants to push that number higher by supporting Microsoft’s Radius cloud-native application platform and so-called “monster VMs” that VMware allows to run up to 960 vCPUs and 16TB of memory across several physical hosts. One of VMware’s crown jewels is vMotion, the tech that moves VMs from one cluster to another with almost no downtime. Memory tiering currently supports vMotion, but Morera said he wants to improve performance by solving some unidentified physics problems. Memory tiering is built into VMware’s ESX hypervisor. Using it requires NVMe drives capable of at least 100,000 writes per second, and 7,300 TB of writes across their lifetime. Another metric commonly applied to NVMe is Drive Writes Per Day (DWPD) – the number of times it is possible to completely overwrite a drive in a single day, without the device dying before its warranty expires. VMware recommends drives with a DWPD of at least three. It’s possible to use memory tiering in a 1:4 ratio – 1 GB of DRAM and 4 GB of NVMe. VMware recommends a 1:1 ratio and claims that it can lead to applications consuming 30 percent fewer CPU cycles, and 40 percent lower cost of ownership by allowing users to “densify” their infrastructure into fewer servers. The Broadcom business unit also argues that adopting memory tiering can improve a server’s performance and therefore extend its working life, making it possible to defer hardware purchases. Company execs argue that’s a win given the long lead times for servers, the high cost of DRAM, and the likelihood that server vendors will need to change quoted prices due to market fluctuations. Of course VMware would say that, because it wants more people to implement VCF. Yet it seems VMware customers and prospects are very interested in memory tiering, as the two sessions on the topic that The Register attended to gather info for this story filled 500-seat conference rooms. ®

Nvidia has for years advanced the idea of an “AI Factory,” and now its partner VMware has created a product with the same name that does most of the same things, but with one vastly important difference: it uses AMD hardware. VMware was an early supporter of Nvidia in the enterprise and long ago created tools to virtualize GPUs. Nvidia needed that help because its hardware was very expensive even before the generative AI boom, so using virtualization to allow higher utilization rates just made sense. The two companies stayed close and in 2023 combined to announce “VMware Private AI Foundation with Nvidia”, a bundle that includes the myriad tools needed to run inference workloads, plus tools to help AI departments automate and manage them. “It simplifies Gen AI deployments for enterprises by offering an intuitive automation tool, deep learning VM images, vector database, and GPU monitoring capabilities,” VMware enthused at the time. A few weeks after the joint announcement, Nvidia started talking about “AI Factories” - essentially a reference architecture that describes all the hardware and software needed to run inference workloads. Nvidia’s AI factories center on its own hardware and software, along with servers from the likes of Dell, Lenovo, and HPE. Those hardware giants brand their implementations “AI factories.” As Nvidia started pushing the AI factory, VMware kept enhancing its Private AI tools that make its flagship Cloud Foundation (VCF) private cloud bundle a good host for AI workloads. On Monday at its VMware Explore conference, the Broadcom virtualization division announced its own version of an AI Factory and said it “streamlines AI infrastructure management by fully automating hardware provisioning, software stack enablement, and end-to-end lifecycle management.” If that sounds a lot like the spiel for the Private AI Foundation, that’s no coincidence because the new offering is a rebranded sequel that adds some useful tech, such as the ability to deploy a model once, and share it securely among users such as tenants or business units. With interest in AI sky-high, this should be welcome as it means organizations won’t need discrete hardware for each of their teams’ AI needs – potentially saving money and putting IT teams in control of a central pool of AI infrastructure instead of trying to identify and rein in shadow AI implementations. VMware has also bundled tools that allow orgs to use AI infrastructure on-prem and in the cloud, and can figure out where to run a job and which model to use to keep costs low. Again, this will be welcome because cloudy AI can rack up huge bills in a hurry, as can using an LLM when a smaller, more specific, and cheaper model can do the job. The Broadcom business unit also offers secure AI sandboxes and governance tools that it says will ensure agents don’t exceed their authority, or access resources you don’t want them to touch. The AI Factory requires VCF, and runs on servers from Cisco, Dell Technologies, Lenovo and Supermicro. Broadcom has also teamed with AMD on a version of the VMware AI Factory that works with AMD Instinct GPUs and the open AMD ROCm software ecosystem. At present, VMware’s AI Factory doesn’t apply to Nvidia’s AI Factories, despite the two sharing some components and intentions. Prashanth Shenoy, VP of product marketing at Broadcom’s VCF Division, told The Register that the VMware AI Factory is an “evolution” of VMware Private AI Foundation with Nvidia. “VMware AI Factory represents a full-stack, automated operational system designed to treat AI token generation as a continuous production pipeline,” he said, pointing out that it supports “multiple accelerator architectures and AI tool chains” and that Nvidia is “our longest-standing GPU vendor partnership.” VMware hasn’t ruled out building an AI Factory for Nvidia and its AI Factories. But for now, the virtualization giant has pinched its long-term partner’s product name for a competing offering that promotes Nvidia’s strongest rival – and all in the name of reducing your AI bills. This is not the only example of strangely enmeshed co-opetition on display here at VMware Explore: On day one of the conference, The Register has already chatted with delegates from Red Hat – which competes with VMware with its OpenShift Virtualization but also partners with it because some people run OpenShift inside VMware – and Microsoft, which offers a cloudy VMware service plus several products that compete directly with Virtzilla. ®