Alex noted that with the release of Heat, OpenStack was encroaching into the PaaS world. This prompted great discussions around this positioning around the Blue Box office:
“DevOps automation is still not PaaS. There is still the need to initiate the DevOps side — it's not deploy and forget. It's build and set conditions and automate resource allocation. It's advanced IaaS management. Cool, but not PaaS. But, in Werner's OpsWorks announcement CloudFormation is specifically NOT even DevOps automation.” – Robert Taylor, Blue Box Solutions Architect
Today, Alex posted an excellent update to his original post titled “OpenStack and its PaaS strategy: a deeper look under the covers” clarifying his position. Alex’s post provides helpful clarification on Mirantis’s positioning in the PaaS market. More importantly, his post is particularly interesting when we consider this week’s announcement of OpenStack’s latest project, Solum.
With the backing of eBay, RedHat, Ubuntu, dotCloud, Cloudsoft and Cumulogic, Solum is designed to combine a full suite of today’s OpenStack services (Heat, Nova, Keystone, Trove, etc) into PaaS functionality with full operational visibility within Horizon. Solum is totally open - it has no committed code and is being designed from the ground up with community involvement. I actually think that’s a really intelligent way to evolve a new project, but as Jack Clark noted in his article on The Register, Solum feels both rushed and a distraction for the OpenStack community.
From my perspective, the problem with Solum is that PaaS and IaaS are about inherently different problem sets, and they should remain that way.
IaaS is about focusing on infrastructure-based services: databases, queues, caches, load balancers. These are the technologies that enable applications to reach scale previously unimaginable without significant investment from an ops perspective. These are also the elements that enable PaaS offerings to flourish.
As I noted in my previous blog post, PaaS is about enabling application delivery – ensuring applications can come to market faster. To me, that means ensuring flexibility in where that application may run.
The reason Cloud Foundry is so interesting and has caught on so quickly is because it truly is IaaS agnostic. Using its Cloud Provider Interface (CPI), Cloud Foundry can be deployed to Amazon Web Services, VMware and OpenStack. Additional cloud drivers can be easily added in (Cloud Stack’s initial driver was introduced this week).
A large part of Cloud Foundry’s success to date can be pinned on the ecosystem it has been able to establish around a core Open Source project. If that sounds familiar, it’s because that’s the genesis and mantra of OpenStack.
OpenStack does not need a PaaS offering to be successful. It needs to focus on delivering core services, and delivering them incredibly well. Let other ecosystems focus on consuming those services and delivering offerings that are truly differentiated. OpenStack has one job. Should OpenStack get distracted from perfecting its IaaS offering by starting down the road to PaaS, it could fail to mature into providing an excellent IaaS.
What if Linus decided that the kernel was too limiting and, like Microsoft, he needed to expand the kernel to include apps and product suites? Would that make Linux more valuable or less?
I could scream this from the mountaintops: OpenStack needs to continue to mature and diversify its suite of infrastructure service. Great progress has been made with Havana, and with the Icehouse release, I hope we’ll see a widened set of tools. I look forward to seeing the evolution of Solum – I don’t doubt it will be an exciting project to watch. I just believe it’s the wrong place for OpenStack to be putting its engineering resources.
It also doesn’t help that all I can think about when I hear Solum is Gollum. Sometimes it's good to not hold on too tight to one's favorite things.
Have you read the news?
Rumor has it this cloud thing is taking off.