In a short time, the edge has become the crucial third leg holding up the IT stool, joining traditional on-premises datacenters and the public clouds. That is not surprising, given the increasingly distributed nature of the enterprise. The cloud, the proliferation of mobile devices, the Internet of Things (IoT), the growing need for more real-time analysis of the massive amounts of data being generated outside of core datacenters and now the promise of high speeds and low latency from 5G networks all play a role in driving demand for more computer and storage capabilities closer to the users and devices created all that data.
But it is still early in the development of the edge and how that will evolve still remains to be seen. But there’s a belief among some system makers that while the edge has particular capability needs and that in many ways the applications running out there will dictate what the infrastructure looks like, there has to be commonality in the infrastructure components from the edge back through the cloud and into the datacenter. Dell EMC officials often talk about the continuum of the IT environment from one point to another.
Supermicro is seeing small but growing demand from enterprises that essentially want to extend out to the edge but make it essentially an extension of their datacenters rather than something separate, according to Michael Clegg, vice president and general manager of 5G and embedded/IoT for the company.
“Some people that have assets out at the edge – the tower companies, the telcos, anybody that has an edge datacenter – and the big datacenter players are starting to say, ‘How can I take my traditional network and push out with these to the edge?’” Clegg tells The Next Platform, noting that such interest dovetails with Supermicro’s strategy of using a building-block model with chips, GPUs, storage and other components that are the same in systems at the edge and in the datacenter. “It’s a design, engineering and packaging problem. Not to trivialize it, but it takes that experience. Typically, you get companies that are either very good at these sorts of industrial computers and you get companies that are good at just traditional servers. What you’re doing with these types of technologies – and this is what’s happening with edge – is you’re taking the technology that’s been in the traditional server and you’re trying to repackage that into almost an industrial server application. You’re starting to bridge those two together and that’s a new expertise that we’re definitely going down that path with.”
The vendor has been growing out its portfolio of offerings for 5G and edge computing environments, including with the single-socket SuperServer E403-9P-FN2T and 1019-FN2T systems, which were rolled out last year. Those are designed to complement datacenter systems like the multi-node BigTwin and high-density SuperBlade and MicroBlade systems that can support virtualized 5G in the network core.
Supermicro this week unveiled the first of what will be a family of small and rugged edge systems aimed at outdoor environments, the IP65 designed for 5G RAN (radio access network), artificial intelligence (AI) inferencing and similar applications that need to be run as close to the user as possible and often in hostile environments.
The systems come with multiple configurations, including options for either Intel Xeon D system-on-a-chip (SoC) or Xeon Scalable CPUs. There also are three PCIe slots that can be used for GPUs – important for AI inferencing at the edge – or field-programmable gate arrays (FPGAs). Storage options include SSDs, M.2 and EDSFF (Enterprise and Data Center SSD Form Factor). The systems are built through Supermicro’s building-block architectural approach.
“We have motherboards that are the core building block,” Clegg says. “Around that we have storage blocks, GPU blocks, power supply blocks and then chasses. It becomes a little bit of a packaging exercise: which combination of these things do I want to do? When you go outdoors you have different constraints, power constraints and most importantly, temperature constraints. Packaging and meeting the extended environmental conditions is the more difficult piece. Usually general-purpose server chips are not designed to over-extended temperatures. Intel does have a class of products that [can handle extreme temperatures], so you pick it out of a smaller pool of processors. For us it’s really going [general purpose] and then doing some custom design as we need it.”
Telcos and other utilities are embracing edge computing environments, as are such industries as smart cities, physical surveillance and retail, according to Clegg. The keys to the edge is not only having compute out there, but also storage, expansion slots for GPU or network accelerators and high levels of performance. Supermicro started out with Xeon D in its edge devices, but customers pushed higher performance and more scalable solutions that could handle datacenter-type applications.
5G Is The Driver
The ramp to 5G is a key driver behind the edge and Supermicro’s efforts behind the IP65 systems. For most 5G environments, the goal is to separate the software from the underlying hardware system, essentially running the applications on general-purpose systems. Supermicro has been working with Intel on its FlexRAN initiative, where a hardware system is made from off-the-shelf components and run what had been networking tasks on dedicated boxes on the commodity hardware, Clegg says.
For vendors like Supermicro, there are essentially two parts of 5G networks. The first is the core network, which is going cloud-native, adopting a lot of virtualization on general-purpose servers and running everything on top of it as a software stack. The other is the radio, which is pinned in by such constraints as digital signal processing. It can be run in a controlled environment like a core or regional datacenter, but in rural areas it needs to be moved to the edge, closer to where bay stations are located.
“That’s the genesis [of the new system],” Clegg says. “Basically, 5G is moving into general-purpose servers to put product out in the field. What we had developed is we had a distribution unit that was designed to go into an edge datacenter or a regional datacenter, but we were getting requests from our customers to put that essentially up close to the tower itself. That’s what drove this development initially. We have taken one of our servers and repackaged that [and] put it inside the IP-65 boxes, a complete outdoor system.”
5G holds the promise of 1,000 times more bandwidth than 4G and LTE, a tenth of the latency, and the ability to support millions of connected devices per mile, a significant capability in the era of IoT. It also will enable such technologies as AI and machine learning, data processing and analytics and virtual reality at the edge and will be crucial in the evolution of such new markets as autonomous vehicles.
“There’s the 5G network itself and then there’s the applications that are going to utilize the 5G network,” Clegg says. “Because the 5G network itself has become virtualized and cloud-native, you often get these two spoken about as if they’re one. The one is enabling the other, though. The work that’s been done in 5G to create these virtualized x86 type platforms, in some cases operators are saying, ‘I will use the same hardware. I will put a unit at the edge of my network – a regional datacenter – and I will run network services on that unit because that’s just general-purpose computing. If it adds extra horsepower, I can also spin an application container up on that same unit.”
What drives 5G is low latency, he says, and low latency drives compute towards the edge.
“That’s probably the catalyst piece of 5G,” Clegg says. “As people think about why 5G is very different from 4G, it’s about the idea of ultra-low latency, sort of the hidden theme in 5G. But the other factor of 5G itself is a virtualized architecture that’s enabling people to really adopt these different compute models. You can probably see some dramatic capex savings down the road as we do some resource shaving at these edge nodes. That’s going to be the other catalyst that doesn’t exist today.”