Today we're announcing an exciting new partnership with Arm to help accelerate adoption of a new wave of Armv8-A based processors in the datacenter.
The Works On Arm partnership, which combines hardware and software innovators and more than a dash of ecosystem outreach, is exactly the kind of work we believe is needed to build the next wave of infrastructure innovation.
Here's a bit of the story.
Word Hunting Future of Arm in the Datacenter
As the resident “Arm” expert at Packet, a colleague recently asked me to pick out a few words that would “make it or break it” for Arm in the datacenter.
To be honest, since I spend my days immersed in the muck of ecosystem development, I was a bit stumped. Four words? Surely it was more complex than that! So I turned the tables on Jacob Smith and asked him to show his cards first. His hand held the following bets: Docker, Go, Android and Kubernetes.
Good bets for sure! Docker, Go and Kubernetes all have massive importance to modern day cloud, and with billions of devices in a mobile-first world, what is better than Android on Arm? There is no doubt in my mind that these four words need to be at the tip of the tongue of every Arm-related executive or fanboy, such is their importance to the success of anyone doing business in the cloud over the next decade.
But after chewing on it for a few days I came back with a completely different set of words - and while I’m not a betting man, I think mine could ultimately be more important to the success of the Arm ecosystem.
Betting on Specialization
Before I unveil my four "game-changing" words, a bit of backstory on why we think Arm even matters. Basically, why should we care if Arm succeeds in becoming a major (or even minor) player in the datacenter chip business?
Well it ends up that at Packet, we have some pretty strong opinions about the future of infrastructure!
One of the most important drivers / trends is a move we see towards greater specialization at all levels of the stack, but especially hardware and silicon. As I.T. applications (databases, CRM’s, etc) are supplanted by new, immersive experiences (virtual reality, autonomy, connected devices, machine learning enabled everything) the amount of computing power is going to explode. We’re still in the very early stages of this trend, and yet it seems as if Google, Microsoft and Amazon - let alone the rest of us - simply can’t build capacity fast enough!
The need for computing power will only accelerate, but counting on Moore's law seems foolish at this point. Instead, as computing power grows, it will need to become vastly more efficient at handling the most important and resource-intensive workloads. These workloads will be so big, and so important that specialization will drive efficiencies far beyond what Moore’s law has enabled in the past. Look around and you'll see giants like Google (Tensorflow Processing Unit) and Microsoft (Project Brainwave) investing in specialty hardware.
Now, when you get to specialization in the processor or device business, you quickly find yourself deeply embedded (pun intended!) in the Arm ecosystem. Due to its structure, Arm licenses its core technologies to dozens or even hundreds of partners - who then customize products around their particular use cases, market needs, related technology, etc. While the universe isn’t unlimited, it is amazing to see the diversity (and the agility) of the Arm-powered ecosystem.
With this in mind, Packet sees a big potential role for Arm in the datacenters of tomorrow. But with just a fraction of the marketshare today, how will it get from here to there? What are the opportunities and risks?
Strength as Weakness
Sometimes our biggest strength is also our biggest weakness, and with Arm in the datacenter I think that is definitely true.
The diversity of the Arm ecosystem - which gives it a natural advantage for quickly creating specialized processors for emerging applications and workloads - has the potential to run headlong into the well-oiled x86 cloud server ecosystem dominated by Intel. In short, most things “just work” when it comes to x86, so bringing a new architecture into the picture - especially one as diverse and dynamic as Arm - is a pretty big effort.
It is an even bigger effort when software is moving as fast as it is today. Just look at those communities that Jacob mentioned: they are moving at lightning speed while looking toward the future. Gazing down and dealing with the hardware under their feet isn’t a natural reaction.
My Four Words for Arm in the Datacenter
Drum roll please! As you can see below, I've chosen ideals as much as words, and stayed away from workloads and technologies.
- Collaboration - At the heart of ecosystem development is a strong sense of partnership, spanning hardware designers, manufacturers, enablers like Packet and dozens of open source and commercial software projects.
- Access - If you can't test, play, build and repeat with automation...you're not going to get very far in today's cloud-centric world.
- Optimization - It's no longer good enough to be "cheaper" and I don't think we're at the point yet where you can steal a substantial amount of datacenter market share on efficiency. Instead, you have to be able to win at key workloads, like those written in Go, run in Docker, and orchestrated by Kubernetes. Optimizing for "what's next" is a privilege that only new entrants to a market can usually afford.
- Adoption - Nothing more than good old fashion sales work here, but with the goal of gaining the early adopters (think "Crossing the Chasm") who will be your advocates, guides, gurus, and references along the way.
One could argue that these words apply to any ecosystem development effort. Probably that is true. However, I think they are especially relevant for the Arm ecosytem.
Announcing the Works on Arm Partnership
Today we're taking a major step forward with our partners at Arm, as well as key ecosystem players Cavium and Qualcomm Datacenter Technologies, to hit on each of these points.
We're actively engaged with the Arm partnership to help set priorities for investment and to share information. We're making available over 100 servers to open source and commercial projects for testing and CI/CD. And we're partnering up with software innovators from Elastic, Open Faas, and Hyper.sh to encourage adoption while helping Arm SoC manufacturers optimize new systems in a real world setting.
So, join us! Request access to resources, help us find issues and opportunities, and let's get to work.