Juniper Networks went all in on data processing units (DPUs) at Nvidia’s GTC event this week where the networking giant previewed its Edge Services Platform.

DPUs — sometimes called smartNICs — are a hardware accelerator tasked with offloading input/output intensive tasks, like networking, security, or storage functionality, from the host CPU. DPUs present a unique opportunity to distribute networking and security services, like containerized firewalls or layer-7 proxies, to every server in the data center, explained Juniper CTO Raj Yavatkar, in a recent blog post.

“These smartNICs can enable a scale-out model so functions like distributed firewalls and virtual routing and tunneling along with load balancing can be deployed and managed across devices,” he wrote.

Nvidia’s BlueField DPUs are just one of several available from vendors like Marvell, Xilinx, Intel, and Fungible.

However, this diversity of DPUs introduces operational challenges when it comes to deployment and management, Yavatkar explained.

“For deploying, operating, and managing SmartNICs accelerating networking and security applications, a logically centralized SDN-like management model becomes critical,” he wrote.

It’s a challenge analysts have cautioned early adopters about in the past. They are “not a drop-in solution to replace standard Ethernet adapters,” and require extensive software development to adapt them to the user’s infrastructure, Dell’Oro Group analyst Baron Fung said in an earlier interview.

Juniper's Edge Services Platform addresses this challenge with lifecycle management, observability, and troubleshooting services for DPU-accelerated, cloud-native applications, and network services. The platform provides a northbound API that enables network fabric managers — like Juniper’s Apstra intent-based networking platform — to automate configuration and deployment.

This, Yavatkar claims, enables data center operators to avoid “islands of network management.”

Juniper is the latest vendor to throw its weight behind DPUs and more specifically Nvidia’s BlueField products.

VMware’s Project Monterey, which entered early access this summer, provides a unified management plane for orchestrating compute resources using DPUs, like Nvidia’s BlueField-2. In essence, this allows users to stitch together virtual machines (VMs) from resources distributed across the data center rather than being limited to a single server.

On the security front, Palo Alto Networks' this summer announced a virtual next-generation firewall designed specifically for Nvidia’s BlueField DPUs. The security vendor claims its VM-Series Firewalls can achieve throughputs close to 100 Gb/s in most use cases, roughly a fivefold increase over running the same firewall on a CPU.

According to Palo Alto Networks, the offering directly addresses the challenges faced by enterprises and network operators as they build out “cloud-like” data centers.

According to Gartner, DPU adoption is growing rapidly, with one in three NICs expected to be a DPU by 2023. To meet demand, vendors have flooded the market with ever more capable products.

Nvidia claims its upcoming BlueField-3 will offload up to 300 CPU cores worth of I/O tasks when it launches next year.

Similarly, Intel unveiled its next generation of infrastructure processing units — what the chipmaker is calling its DPUs — at the Six Five Summit this spring.

However, DPUs aren’t limited to servers. This fall, Aruba launched a new series of switches powered by Pensando’s DPUs. The vendor claims the appliance will open the door for enterprises to operate, scale, and secure their network infrastructure the same way hyperscalers have for years.

And finally, Fungible has eliminated x86 processors entirely from its storage products, which instead rely on its in-house DPUs.