By Jens Söldner and David Marshall

VMware Explore 2024 in the US, the first in-house trade show since the VMware acquisition by Broadcom, is now in the rear view mirror, but the company still has its sights set on Europe, with the coming Explore event taking place in Barcelona on November 4-7, 2024. The US event witnessed a smaller crowd than perhaps anticipated, as many of us who attended fondly remember the heyday of VMworld 2014 in San Francisco, when 26,000 of our closest friends came together to celebrate VMware and its ecosystem of partners. But the event did not lack in news and product announcements.

On the keynote stage, the new man in charge, Broadcom's CEO and President Hock Tan, clearly articulated his mission to steer VMware in a new direction. In that presentation, he called out VMware's previous management for the path taken: 10 years ago, they fell in love with the public cloud, and year after year, presented the audience with different "shiny objects." The result, according to Tan, is that customers are now suffering from post-traumatic stress syndrome caused by the "three Cs" of the public cloud: Cost, Complexity, and Compliance. Those issues will now be addressed, proposing the modern data center under customer control, with the "private cloud" as the solution.

"My view is: the future of enterprise IT is private-Private Cloud, Private AI, run by your own, privately held data. It's about staying on-prem and in control," Tan told attendees during the keynote. He cited the Barclays CIO survey, which indicated that 83% of companies plan to move workloads back from the public cloud to their own data centers.

To implement this vision for VMware's customers, the company is fully focused on its flagship product, VMware Cloud Foundation. This product bundle includes ESXi, vSphere, vSAN, NSX, and the automation and operations tools vRealize. However, the integration of these products has not always been seamless. Broadcom plans to address this and announced the upcoming version, VCF 9, during the event, though no specific release date was provided.

VCF 9 aims to eliminate different GUIs and inconsistencies like varying tagging functions across different components. The company plans to invest in research and development to unify, improve, and simplify the products.

In addition to the core products, customers can add paid services like load balancers, data services, private AI, and enhanced security features like ransomware protection or disaster recovery. According to Hock Tan, customers will then have everything needed to build a private cloud with a feature set similar to AWS. However, what's still missing for a private cloud is a native S3-compatible object storage service. VMware currently recommends either integrating with MinIO or using professional S3-compatible services like DataCore's Swarm. VMware should address this gap.

Beyond integrating and unifying existing services, VMware plans to introduce further innovations in version 9. For instance, VMware plans to build multi-tenancy directly into the product - previously, VMware Cloud Director was required for this. In the future, vCenter and VCF Automation will provide native cloud-like networks ("VPCs") with firewalling and load-balancing capabilities.

VMware also intends to simplify the import of existing topologies in brownfield environments. Technically, the most impressive feature is likely Advanced Memory Tiering for NVMe, currently available as a tech preview. This feature allows memory from NVMe-based storage devices to be used as an extension of main memory.

VMware employee William Lam provided configuration details on his blog. How long the NVMe devices last under intensive operation remains to be seen, but the feature is definitely intriguing from a technical standpoint.

At the same time as VMware Explore, server manufacturer Lenovo introduced improvements to its ThinkAgile VX650 server series, developed in partnership with VMware. The servers come pre-configured for VMware VCF, and Lenovo's professional service department supports installation into the private cloud and serves as a central contact point for support questions.

Of particular interest are certain models of the Lenovo ThinkAgile VX650 V3 series, which, for the first time in the industry, allow the operation of two NVIDIA Bluefield-2 Data Processing Units (DPUs) simultaneously, either to increase throughput or in redundant operation. With the DPUs supported since vSphere 8, the ESXi hypervisor is offloaded to the ARM-based CPU within the DPU, and the network portion of the DPU accelerates the processing of firewall functions from VMware NSX and VMware vDefend.

In practice, various teething problems have emerged with administration and operation since the introduction of this feature, but Lenovo's dual-DPU functionality can help make tasks like firmware patching more manageable. Dual DPU operation is possible with vSphere 8.0 U3 or newer.

The keynote concluded with a presentation by Chris Wolf, Broadcom's Global Head of Private AI in the VMware Cloud Foundation Division. Within his engaging talk, Wolf highlighted the progress of the "VMware Private AI Foundation with NVIDIA" cooperation program, introduced a year ago.

Based on servers from leading manufacturers, customers can use NVIDIA's machine learning cards and a curated software package with VMware's VCF stack to build a complete private AI environment, ensuring that all data remains within the customer's environment. VMware sees benefits in the rapid setup of such an environment, high performance of virtualized GPUs comparable to bare-metal servers, and in some cases even faster performance due to VMware's advanced scheduler.

Additionally, customers can fully utilize expensive and hard-to-source GPUs within clusters. VMware will also offer "vGPU Profile Utilization," providing direct insights into GPU usage.

New partnerships with various companies in the AI ecosystem and a new "Model Store" based on the Harbor container registry will ensure a high level of control and security over AI and LLM applications.

VMware aims to make the operation of AI workloads cost-effective with new high-availability features for GPU usage ("GPU HA") and support for Ethernet as the backbone for AI clusters.

In the keynote, it was mentioned that the total cost of ownership for such a setup could be one-third of the cost of comparable services in the public cloud, although this is hard to verify quickly. Wolf expressed confidence that VMware's technical expertise in distributed resource management, honed over many years, would benefit customers, particularly in managing GPUs, networks, and memory for all types of applications, whether AI or traditional.

If you weren't able to attend everything, don't worry. The recorded presentations from the conference are available online.

And up next, from November 4-7, VMware will hold the European version of the trade show event at Fira Gran Via in Barcelona.

##

(Images: Thanks to Broadcom Inc.)