F5 Teams with NVIDIA to Streamline AI

Application infrastructure provider F5 has taken its load balancing, traffic steering, policy management, and security functions into the realm of AI via an integration with NVIDIA BlueField-3 DPUs. The result, both vendors say, will improve delivery of AI training and inference across multiple dimensions, including speed, efficiency, environmental savings, and security.

F5’s newly NVIDIA-integrated product, announced this week at NVIDIA’s AI Summit in Mumbai, India, is called BIG-IP Next for Kubernetes. The new product works directly with the NVIDIA BlueField-3 DPUs, which offload tasks such as packet processing, encryption, and compression from CPUs in AI clusters. F5’s BIG-IP Next for Kubernetes integrates with these powerful NVIDIA accelerators to add its application-delivery functions for AI training and inference workloads.

“[F5 BIG-IP Next for Kubernetes is] a Kubernetes-native implementation of F5's BIG-IP platform that handles networking, security, and load balancing workloads, sitting on the demarcation point between the AI cluster and other parts of data centers,” wrote Ahmed Guetari, GM and VP of Products for Service Providers at F5, in a blog post.

Tailored to AI Factories

Guetari adds that as service providers and large enterprises expand their AI workloads, they are building AI factories that encompass vast estates of storage, compute, and networking resources. Ensuring that these resources are used optimally is essential to making adequate return on investment (ROI), not to mention best use of environmental resources such as power and cooling.

F5's integrated solution achieves these goals by reducing the amount of hardware required in AI processing while regulating traffic for optimal performance. It also supports multi-tenancy for environments running many AI workloads, a feature that is also handy for service providers with multiple customers.

“This integration provides customers with enhanced observability, granular control, and optimized performance for their AI workloads across the entire stack, from the hardware acceleration layer to the application interface,” stated Kunal Anand, Chief Technology and AI Officer at F5, in the press release.

Fit for AI in Sovereign Clouds

For its part, NVIDIA stressed two aspects of this announcement in its own blog. First, of course, it underscored the role of NVIDIA BlueField-3 DPUs in accelerating AI workloads. But NVIDIA also stressed that the solution lends security and data privacy to environments requiring government-mandated data handling.

That handling requires top security. By running on the DPUs, BIG-IP Next for Kubernetes can deliver zero trust architecture to AI workloads, including edge firewall, distributed denial-of-service (DDoS) mitigation, API protection, intrusion prevention, encryption, and certificate management. By taking over these capabilities, the DPU saves precious CPU time.

Ahmed Guetari gives an example in his blog to prove his point:

“For example, at its Networking @Scale 2024 event earlier this year, Meta mentioned the training of its open-source learning language model (LLM) Llama 3 was hindered by network latency, which was addressed by tuning hardware-software interactions. This approach increased overall performance by 10%. While 10% may seem like a small gain, for a model that takes months to train, this improvement translates to weeks of saved time.”

The F5 BIG-IP Next for Kubernetes appears to be among the first application infrastructure implementations to integrate directly with NVIDIA BlueField-3 DPUs. It’s also aimed at larger-scale AI factories as opposed to integrating NVIDIA GPUs with content delivery network (CDN) solutions to power AI workloads at the network edge.

Futuriom Take: F5’s BIG-IP Next for Kubernetes brings together F5’s application delivery, granular efficiency, security, and policy management with NVIDIA’s powerful BlueField-3 DPU accelerators. The result is a winning combination to target the efficiency and security of AI factories.