Linux Foundation thinks it can get you interested in smartNICs
Step one: Make them easier to program
The Linux Foundation wants to make data processing units (DPUs) easier to deploy, with the launch of the Open Programmable Infrastructure (OPI) project this week.
The program has already garnered support from several leading chipmakers, systems builders, and software vendors – Nvidia, Intel, Marvell, F5, Keysight, Dell Tech, and Red Hat to name a few – and promises to build an open ecosystem of common software frameworks that can run on any DPU or smartNIC.
SmartNICs, DPUs, IPUs – whatever you prefer to call them – have been used in cloud and hyperscale datacenters for years now. The devices typically feature onboard networking in a PCIe card form factor and are designed to offload and accelerate I/O-intensive processes and virtualization functions that would otherwise consume valuable host CPU resources.
Over the past few years, these devices have been crafted to offload a growing number of infrastructure tasks including storage, networking, and security functions, explained Yan Fisher, OPI lead and Red Hat global evangelist for emerging technologies, in an interview with The Register.
Note, for instance, that VMware's Project Monterey allows the virtualization giant's ESXi hypervisor to run on an Nvidia BlueField DPU or Intel IPU.
"Running management and certain offloading tasks over that fabric just makes a lot of sense," Fisher explained.
- Alibaba Cloud challenges AWS with its own custom smartNIC
- SmartNICs power the cloud, are enterprise datacenters next?
- Marvell CXL roadmap goes all-in on composable infrastructure
- Intel sticks with FPGAs and ASICs for next-gen IPUs
In addition to the performance benefits of offloading these services, DPUs and IPUs promise greater flexibility and security by effectively isolating the host processors from the infrastructure layer, Dan Daly, OPI lead and senior principal engineer at Intel, told The Register.
In other words, the DPU effectively functions as a server in front of the server – allowing the CPU to be dedicated entirely to customer workloads.
However, the independent nature of these devices requires software built to take advantage of them. And according to Fisher, most smartNICs up to this point have relied on proprietary software stacks. Hardly surprising, given the industry's stubborn refusal to agree on what the things are even called.
"That was the fundamental thing that's slowed down the adoption" of smartNICs, he said.
This isn't a problem for large hyperscalers and cloud providers willing to dedicate resources to building custom software around these devices' unique capabilities and requirements. But it poses a challenge for the broader adoption of DPUs in enterprises and other businesses.
According to Fisher, the question that the OPI project aims to solve is "how do we generalize, and how do we free up this technology so it will be consumable by everyone."
To achieve this goal, the project will establish a standard definition for what constitutes a DPU (or IPU or SmartNIC), develop common APIs for how software should interact with these devices, and define vendor-agnostic architectures and frameworks on which accelerated applications can be built.
"You need a mechanism to be able to boot that subsystem, to be able to program it, and then to be able to run applications that are going to implement your infrastructure," Daly explained.
The idea being that, eventually, customers will be able to deploy their software on any OPI-compatible DPU and know for sure it will work out of the box.
"These IPUs and DPUs are on to something in terms of their ability to offer new services to the end user," Daly said. "The only way that we're going to be successful here is if we come together with a platform that makes it easier to adopt." ®