Intel’s smartNICs probably aren’t for you (yet), says Intel – The Register | Region & Cash

Unless you happen to run a cloud or hyperscale data center, Intel’s Infrastructure Processing Units (IPU) probably aren’t for you, at least not yet.

Accelerators like Intel’s Mount Evans chips represent an extreme point on a spectrum. The IPU was built in partnership with Google for their data centers, said Nick McKeown, who leads Intel’s Network and Edge Group The registry.

But rest assured, McKeown thinks IPUs will find their way down before long. “It’s going to be a class of devices that I think we’re going to look back at and say, ‘It’s just as important as the CPU, memory and other accelerators,'” he said.

SmartNICs, IPUs, Data Processing Units (DPUs) or whatever you or the vendor’s marketing department wants to call them, have steadily increased in importance in recent years to extract value from systems by making input/output (I/O)-intensive operations. Common examples are networking, storage, and security applications.

The IPU essentially acts as a co-processor, handling the I/O operations, allowing the CPU to focus on its work. This type of functionality has made these accelerators popular in cloud and hyperscale data centers.

“For cloud providers, IPUs offer a way to physically separate the infrastructure layer from the components running customer code,” McKeown said. “The PCI works much like the DMZ, protecting user code and leased client code.”

In other words, these devices allow cloud providers to manage and orchestrate the resources that make up their services without compromising or compromising the security of the revenue-generating CPU cycles sold to customers.

“Right now the focus is on the super high volume. Because when you develop together, you develop with someone who buys a lot of it,” he said.

Because of this, Mount Evans bleeds out of necessity. However, as the technology finds applications in the market – and McKeown believes it will – he expects to see a spectrum of devices with different core counts and performance targets coming to market from a variety of vendors.

Much of what defines Intel’s IPUs can be reduced to less demanding environments, he explained. So one extreme is the IPU and the other a fixed function NIC. “There are no great unknowns in this space.”

There is still a lot to do with the software. “I think what’s perhaps even more important than the device itself, dare I say, is the software ecosystem that’s growing up to support it,” McKeown said.

SONiC for smartNICs

This is where the Linux Foundation’s Open Programmable Infrastructure (OPI) project comes in.

The project, launched last month in collaboration with Intel, Nvidia, Marvell and other smartNIC heavyweights, promises to build an open ecosystem of shared software frameworks that can run on any DPU, IPU or smartNIC.

The organization aims to address a common problem with today’s smartNICs: Most rely on proprietary software stacks, RedHat’s Yan Fisher previously said The registry. “That was the fundamental thing that slowed the adoption of smartNICs.

And while that hasn’t deterred hyperscalers or cloud providers willing to commit the resources to develop custom software around the unique capabilities of these devices, it poses a challenge to their application in other markets.

In many ways, McKeown compares OPI’s efforts to how SONiC kick-started the switching industry. The open-source network operating system developed by Microsoft has brought network vendors together on a common platform and ultimately fueled innovation, he said. “What we’re trying to do as a community is to create the same magical kind of momentum for the IPU class of products.”

However, the project not only promotes the development of novel smartNIC applications, but also addresses another challenge: interoperability from one vendor to the next or from generation to generation.

“When you run an application on a NIC, you’re not actually running an application, you’re running a set of drivers,” McKeown said. “But now, when you talk about an IPU, you’re running a whole microservices infrastructure; You could run OVS … . You need to know that you can lift that up and pass it on to the next generation without having to start over.”

According to McKeown, as smartNIC adoption increases, there is a risk that some vendors may be tempted to lock their software and hardware behind a walled garden. This type of behavior has “held the gear industry back; it held the switching industry back; it has held back a lot of things in the past.”

If not the cloud, then where?

While applications for IPUs are few and far between outside of hyperscale or cloud data centers, that doesn’t mean there won’t be any in the short term.

“A lot of interest is toward the telecom edge, where you can use a class of devices like that to take those cloud ideas and take them to the edge,” McKeown said.

Whether cloud providers are trying to push their own infrastructure to the edge, or telcos are trying to build their own edge computing offerings, there is still a need for highly programmable mesh networks, for the IPUs are well suited, he says explains.

So while IPUs, DPUs, smartNICs, or whatever the marketers come up with next may not be a requirement for servers just yet, it won’t always be the case. ®

Leave a Comment