The growth rate in the adoption of cloud platforms by businesses is too slow.
That’s the message that was explicitly delivered by Intel today during a press briefing to unveil its newest partnership with cloud platform provider Rackspace.
“Despite the fact that we’ve been seeing a 20 percent-plus growth in the adoption of cloud by large businesses, and despite the fact that most people will say that it’s intuitive that cloud is growing like crazy,” said Intel Data Center Group General Manager Diane Bryant, “we conclude that this transition to cloud is actually not happening fast enough. And there are real impediments to that — impediments for private, public, and hybrid cloud deployment today.”
Making ‘Best’ Better than ‘Good’
Rackspace and Intel have both been staring down serious troubles, specifically with not being entirely on the receiving end of the great wellspring of cloud technology benefits that you read so much about.
Just three-and-a-half years ago, Rackspace appeared insurmountable as both a technological and service leader for cloud platforms. Analysts were seriously doubting that Microsoft could catch up to Rackspace, when it took its first moves towards converting Azure to a general-purpose platform.
And with Rackspace having co-created OpenStack — arguably the most successful open source hybrid cloud infrastructure stack — its place in the public mindset appeared permanent.
Intel, meanwhile, has been struggling to claw out a working formula for itself in CPUs for cloud-based servers. It produces three classes of Xeon processor for the server space, following the traditional good/better/best marketing formula.
The Xeon E3 class is what Intel markets as perfect for the cloud. It burns low power, it’s easy to maintain, and it has fewer bells and whistles. It’s also Intel’s lowest-margin product, so even though hyperscale buyers like Facebook (and others whose server purchases follow the Open Compute Project specifications) are liable to purchase E3s in high volume, Intel doesn’t earn much from those sales.
Intel’s mid-range processor class is Xeon E5, which it markets as perfect for the cloud. It features per-core power management, is built for software-defined networking (SDN), and makes the most of the fastest DDR4 memory modules. Xeon E7, meanwhile, now enables so-called programmable gate arrays that ride alongside the CPU chip on the same die, to custom-configure the processor on-the-fly for hyperscale operations.
E7 is something Intel markets as perfect for the cloud. Intel earns higher margins on E7 because it packs the most custom features. The problem is, hyperscale data centers are less likely to purchase E7s, or servers that host E7s, in high volume because they’re on the wrong side of the price curve.
One exception may be Rackspace, which is struggling to sustain its premium value proposition for private and hybrid cloud infrastructure services, against strong competition from Azure, Amazon, and Google Cloud Platform. Rackspace might be in a position to gamble on buying E5s in volume.
Intel wants more data centers to be like Rackspace. So to that end, it’s helping Rackspace establish an OpenStack Innovation Center: a specialized cloud platform comprised of two 1,000-node clusters, that will be made available to qualifying OpenStack developers free-of-charge.
As Intel Cloud Platforms Group General Manager Jason Waxman confirmed to CMSWire, those clusters will use Xeon E5 processors, not E3.
“We want to make the ‘open source’ in OpenStack enterprise-ready,” Waxman told reporters. Over the next six months, he said, Intel and Rackspace will be co-producing what could perhaps be described as a reference implementation: a buildout for server hardware and software utilizing OpenStack, that both companies would prefer.
Waxman said this implementation will include an improved orchestration system for virtualized platforms, including container platforms like Docker. Intel is a charter member of containerization’s two major consortia, the Open Cloud Initiative and the Cloud Native Computing Foundation.
That means Intel has already invested time, effort, and maybe something else too, in Kubernetes, arguably the leading container orchestration platform in current use, and an open source product stewarded by Google. Kubernetes version 1.0 was just generally released on Tuesday.
Perhaps that’s what Waxman meant by improved orchestration? Could Intel be preparing its own distro of OpenStack?
To that last question, Waxman answered no. To the first one ... a definitive maybe.
Waxman told CMSWire that there’s some interest around the possible integration of Kubernetes and Mesos (he might be interested to learn that’s already happened). But the most popular scheduler for OpenStack today, he continued, is its own native Nova component.
“We want to be able to make sure it can do things like continue to maintain uptime, or do things like live migration,” he continued. “These are things that enterprises are going to require.”
This last point is interesting. Live migration between platforms is possible with OpenStack, but it’s tricky, and depends in large measure upon how well the stack is integrating with the hypervisor of the underlying operating system. In a more perfect world, perhaps the OS could be cut out of the equation, and Nova could interact more directly with the on-board virtualization component of the CPU.
More to the point, the Intel VT component. While even Xeon E3 processors support Intel VT, if Intel can make a case for VT working better on E5s, it can better justify its higher price point for E5.
The cloud has drastically altered the information systems economy, rendering the most high-volume data center server purchases as short-term expenditures. It’s difficult to picture respected companies like Rackspace and Intel having to band together to survive in this new market. But that’s precisely what they’ve done.
Title image by Garrett Carroll.