What's best: managed bare metal, VMs via OpenStack, or containers?
Piston's answer: Why not all three?
Piston, known for its OpenStack solution, is rolling out a new version of its CloudOS product that aims to allow organizations to deploy OpenStack-managed VMs, CloudOS-managed bare metal (for apps like Hadoop), and eventually containers -- all governed within the same framework.
CloudOS 4.0's main new feature is the ability to manage Hadoop (and Spark) installations on bare metal rather than inside OpenStack VMs. Hadoop can run on virtualized infrastructure; VMware, for instance, has made some noise about its ability to do exactly that.
But Jim Morrisroe, Piston CEO, believes it's a mistake to ask enterprises to pick either OpenStack VMs, bare metal, or containers alone when the workloads they run are diverse enough to demand any one approach -- or all of them.
"Cassandra, HDFS, Hadoop, Spark -- they run best when they're not contained in a VM," he said in a phone interview. "Having push-button deployment of any of those on top of the bare metal that CloudOS manages is a huge benefit." Likewise, he noted that some analytics applications work best when in VMs rather than on bare metal.
CloudOS deploys Iocaine Linux (a Linux substrate) across the machines it manages, and runs a multilayered management, networking, compute, and storage fabric on top of that. At the highest level are provisions for deploying and running common applications: Hadoop and its associated technologies, OpenStack, and soon container technology.
There, users have their pick of orchestration technology: Kubernetes, Docker, Mesos, or Swarm. Morrisroe emphasized that customer choice should dictate the options: "I don't think we know what will ultimately be the best orchestration for this stuff," he said, referring not only to container technology but orchestration for products like Hadoop.
"We want our customers to be able to figure out what's best for them," Morrisroe said. "If they need to change it out, they should not have to build another silo of infrastructure, and do it on the fly and with their existing racks of gear."
In theory, providing a choice of container technology is complementary to Docker's workings. This is especially true as Docker shifts to what it describes as a "batteries included but optional" model, where modules for functionality like orchestration and networking come with the product but can be switched out freely. It makes sense for CloudOS to complement this approach, but it's not revolutionary.
Unlike other companies preparing to deal with running containers at scale -- IBM, for instance -- Piston plans to leave questions about security enhancements or containerization technology to the groups most directly responsible for it. "We want to leverage the innovations of those [existing] container orchestration projects," Morrisroe said.
"The containerization and the security enhancements, we want to leave that to those communities and those companies. I think that helps our customers reduce risk, because as this evolves, I don't ever think one size is going to fit all. If there's better ways to orchestrate containers, that's what we want to make sure we're there to catch, the integrations of those toolboxes."
[Addendum: When asked which container technologies would be supported, Morrisroe replied, "Our ultimate goal is to be agnostic to the container technology decisions made by our customers. That means we'll eventually have to support Docker, Rocket, as well as generic LXC containers. As new technologies develop around containers, we will work to support those new technologies by partnering with those developing the new technologies."]