One petabyte-scale storage customer traveled to this week’s Red Hat Summit in Boston with dreams of Ceph and Gluster merging into a single product. He hoped Red Hat would take the best pieces of each and “slam them together.” He suggested the vendor could dub the new creation Red Hat Storage, Red Hat Software-Defined Storage or Red Hat Scalable Storage, or “come up with a fun new name.”
“I was waiting for it, and it just didn’t happen,” said Nicholas Gerasimatos, director of engineering at Fair Isaac Corp. (FICO).
And it isn’t likely to happen soon, if ever, according to the heads of product management for Red Hat’s commercially supported versions of open source Ceph and Gluster software.
“They’re very different, and the way we see Ceph and Gluster is that they’re targeted at different parts of the market,” said Neil Levine, the Red Hat director of product management who laid out Ceph’s long-term roadmap during a conference session this week. “Ceph is Fortune 500, ‘I’m building a huge Amazon-style cloud.’ Gluster is not mid-market, but it’s certainly for customers that have a problem ‘that I need to fix and I don’t have months to set this up; I’ve got like a week.’ ”
Levine said the reason the conversation crops up about combining Ceph and Gluster is “mainly because customers want a file system, which Gluster provides, but then they like the distributed smarts underneath Ceph.”
“It’s something that I don’t think we’re likely to do,” Levine said. “Customers can do it, but it’s not a supported configuration that we’re going to recommend or push. I think if you want our file system, you should use Gluster, and trying to put Ceph underneath it, you’re just giving yourself an operational headache and potentially expense if you’re going to buy those products from us as well.”
Separate communities develop the open source Ceph and GlusterFS projects. Ceph’s community is working on a file system, but CephFS is generally regarded as not ready for enterprise prime time yet. In the meantime, Ceph sees use for block and object storage. Gluster offers file and object capabilities.
“Instead of trying to combine the two products, we will come up with a control plane that makes these two products look consistent,” said Sayan Saha, head of product management for Red Hat Gluster Storage. “Our eventual goal is actually to get rid of the whole concept of Ceph and Gluster.”
Saha said the new control plane could provision, manage, monitor and tune Gluster and Ceph in the same way, “where all you care about is data services for your workloads as opposed to caring about where it is coming from.” Red Hat demonstrated the unified storage management technology at its booth in the conference’s exhibit hall. The company wrote the controller software in the last six or seven months, according to Saha.
“You want virtual block storage. You want file storage. Or, you want object storage. You will be able to come to that controller and request that, and it will be served out to you,” Saha said. “If you choose that you want to do block storage, it will give you Ceph, and then you go to the Ceph provisioning. If you say file, it will say Gluster.”
Red Hat currently recommends Ceph or Gluster based on the workloads the customers intend to run, but there can be overlap between the two products. For instance, Gerasimatos said when Red Hat visited FICO, one group told them to use Gluster, and the other group said Ceph. FICO engineers ultimately chose Ceph and decided to run the free, open source version of Gluster on top of Red Hat Ceph in cases where they needed a file system.
Craig Hadix, a data center architect for a global systems integrator, said Red Hat’s two-product storage strategy can be confusing. He said he would be surprised if Gluster is still around in a year. He thinks it would be smart to “take the feature set that Gluster provides and integrate it into one storage product that has Ceph and Gluster features.”
But, combining the source code from two distinct software applications can be a bear of a project. Just ask NetApp. The company spent years trying to merge the scale-out NAS software from its Spinnaker Networks acquisition with its Data OnTap operating system.
“NetApp does aggregation for eight years,” Saha said. “There’s no product.”