VMware by Broadcom has unveiled a new networking architecture that it says will improve the performance and security of distributed artificial intelligence (AI) — using AI and machine learning (ML) to do so.
VeloRAIN (Robust AI Networking), launched Tuesday at the VMware Explore conference in Barcelona, Spain, will “deliver unprecedented visibility, prioritization, and automation for enterprise networks allowing organizations to operate more efficiently and deliver superior user experiences,” the company said in a statement.
The company said it has identified a need for more intelligent edge networking and computing.
In a survey conducted by Broadcom in October, 57% of respondents cited the driver for edge computing and distributed AI at the edge is poor network connectivity between locations. And, the company said in its The State of the Enterprise Edge report presenting the survey, the top benefits respondents plan to achieve by implementing edge solutions are faster response times for latency-sensitive applications (68%) and improved bandwidth/reduced network congestion (65%).
The report was based on interviews conducted in September and October 2024 with 192 respondents across North America, South America, Africa, Europe, Asia, Australia, and New Zealand, plus VeloCloud Edge telemetry which, the company said, included the more than 5 trillion data points analyzed daily, across more than 300,000 edges and 3,700 gateways, representing more than 25,000 customers in 78 countries.
“Each stage of edge technology evolution is capable of transforming a variety of industries,” the report noted. “The latest stage — the intelligent edge — is on the brink of rapid adoption. Deployment data shows that customers have taken steps to implement an AI-ready infrastructure by adopting cloud-first architecture, optimizing connectivity, and addressing security and compliance requirements. However, as AI workloads at the edge continue to proliferate, enterprises must take this moment to reassess their network infrastructure.”
That’s where VeloRAIN will come in.
“VeloRAIN is the foundation of our AI networking innovation, empowering our entire portfolio to better address the demands of enterprise AI workloads,” said Sanjay Uppal, vice president and general manager, VeloCloud Division, Broadcom, in a statement. “By harnessing the advanced capabilities of VeloRAIN, AI workloads from distributed inferencing and agentic peer-to-peer applications to upstream heavy RAG transactions will see improved application-based QoE [Quality of Experience] and security across all endpoints of the enterprise.”
VeloRAIN capabilities
VeloRAIN’s capabilities will include the detection of AI applications using AI and ML, even if the application traffic is encrypted, to enable quality of service (QoS) and QoE for edge AI applications, the company said. It will offer channel estimation intelligence for wireless connections over cellular or satellite to help enable what Broadcom called “fiber-like QoS in the face of dynamically changing network conditions.”
In addition, new applications will be automatically identified and assigned business priorities to insure that critical applications get the necessary attention without manual intervention.
Dynamic Application-Based Slicing, or DABS, will assure QoE per application across multiple disparate underlying networks, whether they support network layer slicing or not, Uppal said during a media briefing. DABS also includes user profiles, so traffic can be prioritized based on a user’s identity.
“One of the things that we discovered when we were measuring AI applications is that the reason why all this makes sense now, and the importance of getting this done for AI, is because the applications themselves are extremely different,” he said. “And they’re different because the patterns are different. The way that they use the network is different. They’re very bursty. They’re highly latency sensitive.”
Zeus Kerravala, principal analyst at ZK Research, agreed. “For today’s CIO, they need to understand that AI will be the biggest disruptor to networks since the cloud,” he said. “In fact, in a recent survey I did, I asked the question, ‘What are your organization’s biggest network challenges?’ and the top response, ahead of cyber security, was supporting AI workloads.
“One of the most compelling features of VeloRAIN is the ability to recognize and prioritize AI traffic. That’s certainly unique to VeloCloud.”
Another analyst immediately saw some real-world applications for the new offering.
“In some ways, this builds on what Broadcom launched with VCF 9, focusing on connectivity across the entirety of the enterprise,” Matt Kimball, vice president and principal analyst, datacenter compute & storage, at Moor Insights & Strategy, observed. “VeloRAIN is really interesting, and I can see the real-world application right away. The need for secure, dedicated connectivity across the enterprise — from the datacenter to the edge — is critical to making AI at the edge real and viable. This seems to be where Broadcom is has trained its focus. The channel estimation intelligence should be especially useful as it ensures that ever important QoS despite those network performance changes that impact virtually every company.”
John Annand, practice lead at Info-Tech Research Group, also saw the connection to previous announcements, noting that VMware had hinted at VeloRAIN in August at VMware Explore in Las Vegas. He has confidence in Broadcom’s ability to solve networking issues.
“As a long-time infrastructure nerd, I’m somewhat of a professional skeptic, but if anyone could figure out wire speed packet and path optimization that could avoid starving expensive GPUs or muddling their order of operations, causing time consuming restarts of complex AI calculations, I’d put my money on Broadcom,” he said. “Broadcom knows how to solve routing problems, so of course, they’re going to lean into that solution set to help solve the AI over Ethernet problem.”
However, he said, “we’re all meant to think that VeloRAIN is about AI workloads; it’s right there in the name, after all. The immediate practical benefit, however, will be consuming Broadcom-built, VMware product-delivered AI technologies without the need for governance, strategies, and ethics boards to improve network operations and end-user experience. For I&O [infrastructure and operations] leaders looking to get the innovation check mark — yes, this will help future-proof you for an eventual private AI cluster. And maybe that is enough to help you get budget to improve those core network services that are the unsung heroes supporting IT’s reputation!”
Source:: Network World