Computing Is Bigger Than The Datacenter
November 8, 2017 Jeffrey Burt
For several years there has been the ongoing debate about ARM and its future in the datacenter. That debate goes on, but the talk is changing.
At the beginning of the decade, ARM Holdings, the company behind the ARM chip architecture that is now owned by Japanese high-tech conglomerate Softbank, said its low-power system-on-a-chip (SoC) designs were a good alternative to Intel’s dominant Xeon and derivative processors for servers and other hardware at a time when energy efficiency in systems was becoming increasingly important.
Over the years that has been speculation about when ARM-based chips would find a foothold in the datacenter. There has been some success with chip makers like Cavium and Applied Micro, and a broad array of OEMs – including Dell EMC, Hewlett Packard Enterprise, Lenovo, Fujitsu and Cray – all have used ARM chips in some systems to varying degrees, and the rise of top-tier cloud providers like Amazon Web Services, Microsoft Azure, and Google Cloud Platform also have opened up avenues to bringing another architecture other than X86 into their environments. However, Intel with its Skylake Xeon SP chips and a reinvigorated AMD, with its Epyc processors, continue to innovate, and some analysts have been frustrated by what they view as a slow slog by ARM to muscle their way into servers and other datacenter systems.
ARM was bought this year buy IT giant SoftBank for $32 billion, injecting significant resources in terms of R&D and other spending into the company. At last year’s ARM TechCon show, SoftBank officials put a focus on the Internet of Things (IoT) and the expected trillion of devices that will connected to networks around the world, and that message carried through at the recent TechCon show this year in Santa Clara.
Just weeks before, in September, ARM announced the hiring of Drew Henry as its new senior vice president and general manager of the Infrastructure Business Unit, which includes servers and networking. Henry’s resume includes almost eleven years at Nvidia, where he led the company’s GeForce GPU unit, as well as time with SanDisk, Acetti Software, and other vendors. Henry recently sat down with The Next Platform at the recent TechCon show in Santa Clara, where much of the focus was on the IoT and computing at the network edge. The discussion came a day after a Qualcomm talked a little bit more about Centriq, the upcoming ARM-based server chip that the company is aiming at cloud providers that will be formally launched this week.
Henry pointed out that with the cloud and IoT, the nature of computing is evolving fast. It’s happening not only in the datacenter, but also the cloud and the edge, and ARM intends to be in all three places. Below is a transcript of the interview, which has been lightly edited for clarity.
Jeffrey Burt: What is the situation with ARM in the datacenter now and what is its potential?
Drew Henry: The way to think about this is a little bit different than the way you asked this question, and let me walk you through it. ARM is a remarkable company, and I’ve decided this is the place to spend another ten years because we are on the verge of just an amazing ten years in compute. The datacenter as a concept is principally dead. The idea of a centralized datacenter – kind of the large place where everyone comes to for compute cycles – that concept is gone. Computing is becoming distributed now, compute cycles are moving off the CPU and onto the GPU, the TPU, and other types of processing nodes. The physical computing is being done elsewhere. Now as you try to get as close to where the decisions need to be made as possible. This is what’s happening at the edge. And so our center of gravity as a company is that edge. First it was with the mobile phone, and now it’s moved very fast into being strongly positioned in the IoT space.
In the IoT space, the infrastructure is, how do you build the infrastructure that is critical to support this incredibly complex IoT environment. That’s what we’re focused on. That’s why I say, I don’t think about where we sit from the datacenter standpoint, I think about where we sit enabling that to happen. We are really strong in network today, we are really strong in compute technologies today. This new world of a trillion devices on the edge is complicated by the fact that it’s a trillion devices. It’s also complicated by the fact that the software that’s running on these things is microservices-based software. Each one of those microservices in and of itself becomes a node on a network; each microservice has an IP address associated with it. So you’ve got remarkable layers of complexity just with the interconnection of these things. We are really good at that from an infrastructure standpoint.
We’re also really good when you look at all those devices and say, ‘I want to have compute that is supporting all those things.’ We have a remarkable amount of compute IP to support that. What we’re working on is, how do you build infrastructures to support all that? That means, how do you interconnect it all, as physical devices as well as the interconnect software that’s doing all that, and how do you do all the compute necessary to manage the data, to process the data. Whether that compute happens at the bay station, at the downlink of a 5G distribution network, or if it happens back at a large data center, it doesn’t matter to me. We will be at each one of those places.
Jeffrey Burt: So are you saying that over the past year and a half or so that ARM’s view of computing and where it happens has changed?
Drew Henry: No, it hasn’t changed. Our view has always been around being present at the edge. The way we have articulated our position in infrastructure has been a little more towards a datacenter-centric kind of view, and the way we are thinking about it now and moving forward is, how do you build an infrastructure to support that. We will be in the datacenter, we will be in the compute areas, but it’s much more from an edge-in view than a datacenter-out view.
Jeffrey Burt: During Qualcomm’s talk here about Centriq, the focus was on the cloud providers they are targeting, about datacenters. So when you do look at the datacenter, how would you assess ARM’s future in it?
Drew Henry: It’s strong, and what I love about all the work Qualcomm is doing is that they’ve built solutions that have a series of attributes. One of the attributes is the remarkable compute density. The compute density in that power profiles, that’s a remarkably beneficial thing for datacenters. But the second is the recognition that there are certain classes of workloads that are appropriate for the architecture that they’re bringing. Those are the workloads that are more important to the future of computing. They’re important relative to that view I’m talking about, where you have much more distribution of devices, so the machine learning that Qualcomm has built into the Centriq – those workloads that they’re able to do? Those are perfect workloads to be able to help with distributed IoT tech environment.
Jeffrey Burt: ARM has been talking about moving into the datacenter for several years now. It seems as though it’s gotten stuck. So what has changed when you look at ARM’s server prospects?
Drew Henry: I don’t think anything’s changed. It’s hard to enter into any market with a different perspective than the market has. It’s hard, because you’ve got to have – depending on the workloads – the right type of software stack in place. There’s a lot of elements you’ve got to pull together, and the company’s done a great job in these areas.
The second thing related to it is that, when you look at the workloads that are running on servers, those workloads are becoming much, much more I/O-dependent. That I/O dependency is a really strong position for us because we have such a strong position in network processing. When you look at the smartNICs being built around the world, ARM is in a lot of them. We are processing the critical packets of data that are being sent around right there in those datacenters. That is all part of enabling this kind of distributed type of computing I was talking about. It always takes a long time to try to get an entire ecosystem enable to enable that kind of vision. If the vision’s clear, the ecosystem will come.
Drew Henry: The X86 world has always been the X86 world and will always be the X86 world. Our job is not to go off and try to build products that are alternatives to x86. If you just want some other type of x86 processor, buy AMD, which is an outstanding company. What I believe that needs to get built is the compute and infrastructure necessary to support the workloads that are needed for the edge type of world that is being built today. The past has been this datacenter type of view where I’m just saying, ‘I’m going to go off and I’m going to do some type of process of some type of stuff in the datacenter.’ It’s not where compute is going to be in the future. That is an x86-centric world. The ARM-centric world is at the edge.
Jeffrey Burt: But you do have x86 at the edge as well. You have Intel pushing a vision of x86 from the core to the edge and into the cloud – with some presence in the devices. So where is ARM’s advantage in this?
Drew Henry: If people want to use the closed ecosystem that is represented by x86, they can go down that closed ecosystem path, but I just don’t think the next ten years computing is going to be restricted by one particular view. I think people will want to disaggregate and have a different computing stack than is just the x86 computing stack.
Jeffrey Burt: So where is ARM now in this evolving computing environment?
Drew Henry: We have the IP that is critical for this edge-based world, so if you want networking technology, if you want compute technology, it’s all available to you across the entire ecosystem of partners that is our ecosystem. People will have whatever you need. One reason why Apple was so successful was because they were offered the ability to disaggregate the CPU. Intel walked in [and said], ‘Here’s an X86 processor.’ Apple walked in and said, ‘We don’t want to draw the boundaries the way you’ve drawn the boundaries. We want to be able to construct something different than the way you’re proposing. ARM walks in and provides a module that allows Apple to go off and build something that is remarkable, and remarkable not because X86 isn’t a good architecture, but it wasn’t packaged the way that that was beneficial to building phones.
The same thing is happening with IoT devices. IoT devices are not going to be built from a fixed view of the world. IoT devices will be built on something that is much more modular.
Jeffrey Burt: How do the next two or three years play out for ARM?
Drew Henry: It will play out as it’s playing out. The momentum for what we’re providing is just going to continue to grow. The enthusiasm I see from all the OEMs I talk to – everything from operators that are building out the 5G networks to the datacenter to the cloud guys that are building out these public cloud solutions. These guys are all coming to us saying, ‘We want to work closely with you to build out what we need to have for our view of computing.’
Jeffrey Burt: Regarding Qualcomm, what’s been the impact to ARM in having a top-tier chip maker making a push in the datacenter? Other companies like Cavium and Applied Micro also are making datacenter chips, but Qualcomm has much more size.
Drew Henry: Qualcomm is a force in the entire IT industry. They have a remarkable amount of market strength and knowledge, and the fact that they recognize that this world is going to be an edge-based, trillion-device world – and how do you build an infrastructure to support that world – they share that view. They’re going to put servers that are Qualcomm-based servers into the market, there are going to be devices that sit on the edge that are going to be Qualcomm-based devices, all built around a vision that they have that is very much aligned with ours.
It has built upon a growing amount of momentum and support. Microsoft [in March] stood up at a conference and said, ‘We are going to build a significant portion of Azure on ARM.’ It was a public statement. Qualcomm is just yet another major company in the IoT space that has built upon this vision.