My kids are 5 and 7 years old. By the time they are both of “driving age” in 10 years, there likely will be no need for them to actually learn how to drive. While it seemed far-fetched just a few years ago, it no longer is anything but likely that there will be a dominant self-driving car culture within the next decade.
So when my boys are 15 and 17 years old, and there are 30,000 self-driving taxis zipping around NYC, what kind of compute is going to make it all “just work”? Will the Artificial Intelligence and code driving my kids up Broadway at rush hour be powered by a fleet of generic VM’s running 280 miles away in Ashburn, VA? My guess: not a chance.
Designing the Future, Today
It’s good to remember that both AWS and our trusty iPhone just turned ten years old.
If I had asked you in 2007 what the Internet would feel like in 2017, you probably would have had a hard time putting down your Blackberry keyboard to imagine the public cloud dominance we have today. But with software innovation accelerating down the stack and the 21 million professional software developers on our planet today set to grow by 45% in the next 5 years, I bet it’ll look even more changed five years from today. Most developers I know have a habit of hacking away and reinventing things every few years and I suspect this coming cycle will be no different.
Bill Luby, one of the founders of Seaport Capital and an early investor in the internet infrastructure space (and Packet!), regularly reminds me that as a startup with almost no history, we have all the reasons in the world to look towards the future - very few to look at today. We literally exist for the future! Yet the bigger the company we become, the more established the business, it is only natural that we will be weighed down by what’s currently happening or has just happened to us and our market. I love this advice from Bill, as it reminds me that our startup is on borrowed time.
At Packet, we’re taking full advantage of our "youth" and spending as much time thinking about the future as we possibly can.
To do this effectively we constantly ask ourselves: what will the Internet feel like in 5 years? 10 years? How will internet infrastructure, application deployment and delivery, and networks work? What are the necessary primitives that will power those 30,000 self-driving taxis?
Our goal in pushing the timeframe out is to think beyond discussions about Kubernetes 1.7 or the upcoming DockerCon product announcement, and work today to design the infrastructure for tomorrow.
Ping, Power, and Processors
Today we announced the addition of 11 edge locations in cities from Sydney and Singapore to Ashburn and Dallas. This brings our total footprint for on-demand bare metal to 15 locations, effectively putting ping, power and processors across 3 continents and within reach of hundreds of millions of end users. We’ll be adding more pins to the map in places such as Mumbai, Sao Paulo, and London in the coming months.
We did it for more than just pins on the map though. These locations - which feature a single server configuration and Packet's signature networking and automation capabilities - are but the first aspect of our vision to make fundamental infrastructure available at a global scale to developers (and the software they are creating).
You see, we have a secret passion. One that we are committed to being the absolute best at in the world: automating fundamental infrastructure. Basically, hardware of any type or flavor at the end of an API.
Two years back, in August 2015 we deployed the first iteration of that automation in a single datacenter. 8 months later, we deployed it in two more places (on the same day!) before adding Tokyo six months after that. Today we are adding eleven more sites, pushing ourselves and our platform further afield. You can see where this is headed!
Today, you can deploy a diverse set of hardware in one of our four core facilities, and a single config in eleven more locations. As a developer, you can apply your Terraform plan and get a global fleet of bare metal machines for about $7 per hour. This, in and of itself, is pretty darn cool.
But what we’re really proud of is that we can give this to our customers with as little opinion as possible -- allowing nearly every startup or innovator to play with the Internet via BGP, announce their own AS and grow their own network, or reinvent their own operating system substrate with our custom IPXE service.
In the future we expect the number of locations to grow and the diversity of hardware to evolve. But you won't see us opening up 1,000 compute farms the size of football stadiums. Instead, our hundreds or thousands of locations will look a lot more like the hipster specialty food market at the corner of your block, instead of the Sam's Club 30 minutes away next to the freeway.
Four Words to Describe the Future of Compute
While the future is never what you expect, we feel pretty confident that robot taxis will need infrastructure that is pretty different than today's options.
Why? Well, mainly because the current phase of public clouds are really hard to drop close to major cities (where the people and experiences exist), and because they are too generic and therefore expensive and inefficient for big, special workloads. Model it out to any scale and you can see the issues.
Since we are primarily in the “hardware at the end of an API” business and don't have religion about all the rest of the stuff (software, architecture, etc), we've been able to boil our vision down to four words that we think describes the next wave of Infrastructure fabric:
- Localized - Latency-specific factors and legislative / regulatory requirements will drive infrastructure to be more local and customized to the environment it is trying to serve.
- Distributed - Physics come into play at some point, so in addition to political and business concerns, workload will need to move closer to the eyeballs and get far more power efficient. It’ll be highly distributed or must have the ability to do so. Just think about your mobile phone.
- Untrusted - When you start putting expensive and important stuff like hardware in lots of places you learn that you can't trust it and all the SAS70 audits in the world won’t help you when your workload is running in a city you’ve never even heard of. There are no man traps at cellphone towers bolted on to some substation outside Mexico City. As such, software and hardware will have to work together to trust each other - or not. Better get used to starting with an "untrusted" state.
- Specialized - Commodity processors are great for general workloads, but take a look at Google, Microsoft and others and you'll see that the silicon is getting mighty specialized. We expect this to accelerate, with more innovation and performance leaps happening at the hardware layer. So the expertise will move from deploying 100,000 of the same thing in one place to some far more diverse and disaggregated iteration.
Packet’s mission is to Build a Better Internet -- specifically by enabling the hardware substrate of the future for a (millennial) generation that desperately wants to tinker on a global scale. It is our hope that moving from four to 15 global locations is a first step in creating an infrastructure petri dish that will power those next big things.
In case it wasn’t clear, we’re just getting started here. You can expect more fundamental primitives to show up across our platform API’s in the near future, particularly around pricing, capacity management, and reservation models. These are the tools in the toolbox that we need your help battle testing and refining.
Most of all, we can’t wait to see what you all create. We’ll be listening hard and doing our best to keep our eyes on the next. May the games begin!