Cumulus Linux 2.5, designed for white-box data center hardware, helps Hadoop, VMware, and OpenStack users ramp up their networking.

Cumulus Networks‘ vision for data center networking involves network switches with a single, commodified, and wholly open operating system that runs on generic hardware: Linux. It is pursuing this vision with the new Cumulus Linux 2.5, created to be easier to deploy in conjunction with the solution stacks the company has seen most commonly assembled, such as as VMware.

Cumulus Linux’s advantages go beyond replacing proprietary router OSes like Cisco IOS or being a cost-saving measure that can be deployed on generic hardware. Since the OS is Linux, conventional data center tools like Puppet or Chef are compatible with Cumulus devices and can be used to extend the automation to the networking plane.

Cumulus Linux 2.5 was announced today and is scheduled for general availability later in the year. Its new features center on support for the new Layer 2 network architectures, while retaining interoperability with existing Cisco hardware. But the company claims the real advances have to do with adapting to the “design patterns” seen by Cumulus engineers in the modern data center.

In a conference call, CEO and co-founder J.R. Rivers said the real-world deployment and maintenance of common application stacks drives much of Cumulus’s viewpoint. The big three were VMware vSphere, Hadoop (and other big data solutions), and OpenStack.

“[Supporting those configurations] was not about our partnerships,” Rivers noted, “but rather about what our customers had emerging in their data centers.” Cumulus consulted with Mirantis to garner more details about OpenStack in the field; it did likewise with Hortonworks for Hadoop.

To that end, Cumulus created full deployment solutions for each scenario — deployment tools and step-by-step guides to using Cumulus Linux in each environment, either for a full-blown deployment or just a trial. (Cumulus makes a 30-day evaluation version of the OS available for free.)

Rivers described a key difference in the way the network designs of the bigger outfits Cumulus surveyed — the “megascales” — shaped up against more conventional enterprises. The megascales “tend to use IP networks with network virtualization overlays, kind of like VMware NSX.” Sometimes it’s a homegrown version, he stated, while at other times it’s VMware NSX itself.

“Many enterprises, though,” he added, “aren’t quite comfortable with the network virtualization and overlay part yet. They still want to rely on some kind of old-school, legacy tools like multichassis LAG and a lot of VLANs.” Those organizations want to progress to a more modern, IP-oriented model, he said, but need help getting from their current legacy setup to a modern one.

While use of OpenStack was featured as one of the major scenarios that Cumulus wanted to support, Eucalyptus — now an HP product — didn’t make that cut. (“We’ve run into almost everyone else but them,” Rivers admitted, “but I hear about them all the time; maybe we just keep missing each other.”)
Another more major omission from the list of solution stacks was anything Microsoft-related. In its experience, Rivers found that Cumulus hadn’t found anyone using the stand-alone Microsoft products. “Azure, yes,” he added, “but we’re not running across people standing up much of that themselves.”