Has Linux got OpenStack licked? The Vanilla 'Plus' strategy
Days of froth and noise inside the big tent are over
No man is an island, nor is OpenStack removed from the fates of two of its – until recently – best-known names.
Hewlett-Packard Enterprise was once the largest single contributor to OpenStack. Now, HPE is getting rid of its OpenStack engineers as it turns over responsibility for its OpenStack work to MicroFocus under a deal announced in September.
The ascent of OpenStack-consulting wunderkind Mirantis experienced a hiccup with its decision to unload 100 OpenStackers and reallocate another 100. Mirantis is understood to have let go devs working on the Solum platform-as-a-service and Savanna Hadoop-as-a-service projects it started in 2013.
It has dismantled a team working on a project called Workloads for PaaS and redirected engineers to work on Kubernetes while it's dramatically re-architecting its Fuel OpenStack management project for a new release in the first quarter of 2017.
A Mirantis spokesperson stressed that Mirantis is not abandoning OpenStack or the OpenStack distribution business.
HPE said its offloading of OpenStack engineers is part of "previously announced restructuring". It announced HPE Helion OpenStack 4.0 and Stackato 4.0, for PaaS using Cloud Foundry, in October.
The firm told The Reg it is finalising its partnership with MicroFocus as its "strategic technology partner" for OpenStack and Cloud Foundry. Circumstantially, at least, this is a bad day for the nearly decade-old project supposedly offering an open alternative to Amazon's AWS and Microsoft's Azure.
It would be a crude simplification to attach too much significance to events, given HPE and Mirantis are separate companies with different pressures. And yet the changes are important.
There's an increasing sense that the fast-growing, anything-goes days are finished and a shakeout is in progress, with more exits expected. Going are projects, installers and monitoring systems, and boxed and branded versions of OpenStack that grew like weeds in the last few years. What's likely to be left will be the Linux distros offering vanilla installs with plug-ins.
Seven years in, OpenStack has found its level with deployments settling into a familiar pattern: 95 per cent using Nova, Neutron, Cinder, Keystone, and Glance. Around half, 50-60 per cent, run Metal or Ironic and Heat, pointing to a relatively rapid adoption of OpenStack for micro services and with containers.
Jonathan Bryce, OpenStack Foundation executive director, reckons a misconception had grown around what he called the "big tent" nature of OpenStack – that the projects that arrived there were destined for traction.
"Some misunderstood the big tent as all projects are equally important," he told The Reg. "It was really about upstream management of software development and a place for people to try out ideas and see if they got traction, and see if developers and users came. So, naturally, some of those projects have gone away and others will grow. [OpenStack] has winnowed out some and it will winnow out some more."
Canonical and Ubuntu founder Mark Shuttleworth told The Reg that the big tent had contained "froth and noise" until recently. The reason being a rush of legacy tech firms and startups to whom OpenStack was the answer.
"When PaaS became fashionable a bunch of people announced OpenStack PaaS even though they had no background – but how could they lose?" he said. "Now people are realising if they can't make it in the real world, coming from OpenStack and announcing an OpenStack platform doesn't make you a winner.
"For a long time OpenStack was a combination of hubris inside the community because it was so obviously the only game in town – an open forum for infrastructure and service and so many companies struggled to imagine a cloud future for themselves, creating bridges back to legacy products."
According to Shuttleworth, there's now a broad acceptance about what OpenStack is. It is infrastructure as a service, says Canonical's man.
Arguably, few were as responsible for churning the OpenStack froth as HPE. As HP, the firm was to become a provider of public and private cloud services for everybody. Over time, as HP became HPE through the seemingly never-ending contortions of Meg Whitman's turnaround, the vision shrank.
HP in 2014 overtook Red Hat to become the single largest vendor committing code to OpenStack – 19 per cent – versus Red Hat's 17 per cent. It followed a plan to spend $1bn on research and development of OpenStack to build its Helion over two years. HP invested heavily in the management layer.
Today, HPE is not only not an OpenStack ISP, it has also outsourced its work on OpenStack to MicroFocus. Inevitably, HPE maintains that it's still committed.
Mirantis was among a crop of startups capitalising on a sudden shortage of skills. Through consulting and a branded distro, OpenStack offered to get you up and running. It was a fix for the times, but what next? What after the customer has installed?
Before the changes at HPE and Mirantis, the story of the times had been one of big vendors snapping up private OpenStack cloud providers for their own: Oracle took Nimbula in 2013, Cisco took Metacloud in 2014, and IBM grabbed Blue Box in 2015.
Bryce reckoned the proliferation of OpenStack distros came at the wrong time, causing confusion. "We have seen loads of distributions, too many," he said. "It was confusing because how do you pick Ubuntu, Red Hat, SUSE, Mirantis, Metacloud, Blue Box? People were barely getting into OpenStack and had a massive list of options to pick through. It wasn't helpful to adoption."
As the froth pops and the noise fades, it's the Linux distros who look likely to emerge as the Darwinian survivors. Fujitsu, for example, already uses Linux shop SUSE's Cloud as the OpenStack reference architecture for its servers, storage, and professional services.
"Fujitsu appears to be saying: 'We don't want to just develop our own distro, we will partner with SUSE who has experience with Linux and let them do that and we will deliver it to market with our hardware and services.' I believe that is what we will see," Bryce said.
The Linux distros are well positioned thanks to their experience in converting upstream Linux code into software that can be easily consumed by the ordinary user. The Linux firms you see today are the survivors of the shakeout long ago of the initial rush into Linux.
The era of boxed and branded OpenStack as a product category has quickly passed, commoditised as OpenStack has consolidated around the same installations – Nova, Neutron, Cinder, Keystone, Glance.
That makes OpenStack a difficult place for big vendors like HPE or startups new to managing code streams. But it's an opportunity for the Linux firms, according to Shuttleworth, who sees Canonical's job as providing vanilla OpenStack plus extensions.
He points to EMC's ScaleIO Block Storage driver, which creates elastic virtual SANs at a fraction of the cost and complexity of traditional SANs. It's integrated with OpenStack's Nova and Cinder, and works with Canonical's JuJu charms.
"There are integration points that are interesting – if you want to buy vanilla OpenStack but you want to weave in some special workloads that won't be an add-on to the standard core of OpenStack,” he said. "I think the distros will shake out in favour of broadly vanilla plus those extensions."
As the field clears, are we likely to see a re-run of the on-prem battle between Linux distros, a spat that's come to be largely defined as Ubuntu versus Red Hat?
RHEL dominated enterprise servers but lost ground in the pre-cloud ISP world to CentOS, with Canonical now claiming that Ubuntu runs in 70 per cent of public cloud workloads and 55 per cent of OpenStack. It's an assertion that Shuttleworth reckons helps underling Ubuntu's credentials as a Linux platform for vanilla OpenStack "plus".
Now HPE's shoving off, Red Hat is once more the single largest corporate contributor of code to OpenStack.
"Always Red Hat is the competition," Shuttleworth said. "A lot of folks went with Red Hat because it was open source but a lot of folks have realised the challenge with OpenStack is how to upgrade it and expand it – those are things we ingested long before anyone else.
"The fact we have so much on the public cloud is a big win for us. Mentally it doesn't seem like a big leap for them to buy from us where they would have bought from Red Hat." ®