Over the past year, we at Joyent have been weaving some threads that may have at times seemed disconnected: last fall, we developed a prototype of DTrace-based cloud instrumentation as part of hosting the first Node Knockout competition; in the spring, we launched our no.de service with full-fledged real-time cloud analytics; and several weeks ago, we announced the public availability of SmartOS, the first operating system to combine ZFS, DTrace, OS-based virtualization and hardware virtualization in the same operating system kernel. These activities may have seemed disjoint, but there was in fact a unifying theme: our expansion at Joyent from a public cloud operator to a software company — from a company that operates cloud software to one that also makes software for cloud operators.
This transition — from service to product — is an arduous one, for the simple reason that a service can make up for any shortcomings with operational heroics that are denied to a product. Some of these gaps may appear superficial (e.g., documentation), but many are deep and architectural; that one is unable to throw meat at a system changes many fundamental engineering decisions. (It should be said that some products don’t internalize these differences and manage to survive despite their labor-intensive shortcomings, but this is thanks to the caulking called “professional services” — and is usually greased by an enterprise software sales rep with an easy rapport and a hell of a golf handicap.) As the constraints of being a product affect engineering decisions, so do the constraints of operating as a service; if a product has no understanding or empathy for how it will be deployed into production, it will be at best difficult to operate, and at worst unacceptable for operation.
So at Joyent, we have long believed that we need both a product that can be used to provide a kick-ass service and a kick-ass service to prove to ourselves that we hit the mark. While our work over the past year has been to develop that product — SmartDataCenter — we have all known that we wouldn’t be done until that product was itself operating not just emerging services like no.de, but also the larger Joyent public cloud, and over the summer we began to plan a broad-based stand-up of SmartDataCenter. With our SmartOS-based KVM solid and on schedule, and with the substantial enabling work up the stack falling into place, we set for ourselves an incredibly ambitious target: to not only ship this revolutionary new version of the software, but to also — at the same time — stand up that product in production in the Joyent public cloud. Any engineer’s natural conservatism would resist such a plan — conflating shipping a product with a production deployment feels chancey at best and foolish at worst — but we saw an opportunity to summon a singular, company-wide focus that would force us to deliver a much better product, and would give us and our software customers the satisfaction of knowing that SmartDataCenter as shipped runs Joyent. More viscerally, we were simply too hot on getting SmartOS-based hardware virtualization in production — and oh, those DTrace-based KVM analytics! — to contemplate any path less aggressive.
Of course, it wasn’t easy; there were many intense late-night Jabber sessions as we chased bugs and implemented solutions. But as time went on — and as we cut release candidates and began to pound on the product in staging envirionments — it became clear that the product was converging, and a giddy optimism began to spread. Especially for our operations team, many of whom are long-time Joyent vets, this was a special kind of homecoming, as they saw their vision and wisdom embodied in the design decisions of the product. This sentiment was epitomized by one of the senior members of the Joyent public cloud operations team — an engineer who has tremendous respect in the company, and who is known for his candor — as he wrote:
It seems the past ideas of a better tomorrow have certainly come to light. I think I speak for all of the JPC when I say “Bravo” to the dev team for a solid version of the software. We have many rivers (hopefully small creeks) to cross but I am ultimately confident in what I have seen and deployed so far. It’s incredible to see a unified convergence on what has been the Joyent dream since I have been here.
For us in development, these words carried tremendous weight; we had endeavored to build the product that our team would want to deploy and operate, and it was singularly inspiring to see them so enthusiastic about it! (Though I will confess that this kind of overwhelmingly positive sentiment from such a battle-hardened operations engineer is so rare as to almost make me nervous — and yes, I am knocking on wood as I write this.)
With that, I’m pleased to announce that the (SmartDataCenter-based) Joyent public cloud is open for business! You can provision (via web portal or API) either virtual OS instances (SmartMachines) or full hardware-virtual VMs; they’ll provision like a rocket (ZFS clones FTW!), and you’ll get a high-performing, highly reliable environment with world-beating real-time analytics. And should you wish to stand up your own cloud with the same advantages — either a private cloud within your enterprise or a public cloud to serve your customers — know that you can own and operate your very own SmartDataCenter!