Joyent, a cloud provider and cloud software vendor I’ve talked about a little in the past, has released a new version of its SmartOS stack (Joyent7) that is making the distinction between the Joyent approach and that of traditional clouds like Amazon or OpenStack a bit clearer. Not clear enough, though, I think. There’s still a bit of an articulation issue here.
Most clouds today are built on hypervisor technology, and the cloud software is essentially a management front-end that allocates machine images to VM instances. This approach is usually characterized by a lot of “hypervisor-agnostic” comments in the marketing material. The benefit is that users of virtualization or those doing server consolidation in an attempt to get out of the hole of past IT planning disorder can bring things home more easily. The problem is that it’s not really the cloud at all.
I think “the cloud” is a new IT paradigm built on a set of services that collectively form something like a cloud-distributed OS. The problem with the traditional hypervisor clouds is that they don’t really have any platform for services. Sure you can build a service, stick it in a VM, and let it live in a virtualized instance, but you’re replicating all the middleware and OS overhead for every one you deploy, which is bad enough for IaaS but fatal if you’re trying to build a cohesive cloud that supports cloud-native apps, which is where we’re heading if there’s any substance to the cloud at all.
Joyent’s approach has been to take what’s arguably the best real-time OS that ever was, Solaris, and use one of its open-source forms (Illumos) to write a true cloud OS that runs on the bare metal. The KVM hypervisor lets the system run machine images like IaaS services in “zones” that are fairly well isolated (not quite as well as hypervisor-on-hardware systems would isolate them) but that are elastic in resource allocation so they’re not wasteful. Native cloud apps can be written on the platform directly, but Joyent is also a key backer of node.js, a server platform programmed in Javascript. Node.js can be used to build a bunch of cloud tools and services, and so could be said to be one of if not THE first cloud languages. SmartOS also inherits the ZFS file system that’s the most scalable of all the modern file frameworks for OSs.
In IaaS application, zone-hosting KVM and IaaS machine images is much less resource-intensive because it doesn’t waste resources on fixed allocations. Zones can also contain native applications and the management of the two are essentially the same, and best of all the native or node.js apps are running right on the cloud OS, so no matter how many of them there are, you don’t have replication of a bunch of platform software with each image to clog things up.
You can get this all in three forms; a public cloud service based on the software, a server platform (SmartOS) and a data center complex with a bunch of added integration tools (SmartDatatCenter). With the new Joyent7 version it should be possible to create hybrid clouds nearly seamlessly and also to create federations of providers (Joyent’s Global Cloud Network is such a federation).
All this good stuff would make you wonder why Joyent isn’t a household word, and even more so when you reflect that they are the leading cloud platform mentioned by operators in our surveys (and have been for a year). The reason is that the company, like all tech startups, is hopelessly mired in techobabble and can’t get the message out in digestible form. This year they got a major funding infusion and the new investors are gradually sorting out the organizational issues. They’ve just added a new CEO, Henry Wasik, who came from Force10, successfully sold to Dell. He had a good reputation as a tech leader, and obviously a better one as a guy who can flip a company for a profit. Likely the new investors hope for similar heady results, but to get them it’s clear that Joyent is going to have to do a heck of a lot more to merchandize itself. It’s probably the most future-directed of all the cloud platforms, which means that in my view it’s less shooting at Amazon’s narrow EC2 concept than at its stealth, evolving, AWS API-centric evolution to a service-based cloud. Saying that would be a good place to start your image-buffing, Joyent. IaaS is a transition strategy only, and winning there just hangs you up short of the goal line.