Mesosphere OS manages entire data centers with one click

Mesosphere is best known for its work on the Apache Mesos project at the core of its "data center operating system," Mesosphere DCOS, which can make a whole data center's worth of systems act like a single system without changing OSes or hardware.

Now Mesosphere is releasing DCOS for general availability in two editions: a free community edition and a for-pay enterprise version, both aimed at making the most of public cloud resources.

Run one, run all, run anywhere

DCOS was developed as a way to manage with a few terse command-line operations both running applications and the Linux systems they're deployed on. It doesn't require a rip-and-replace for existing deployments of Linux; rather, it manages existing Linux systems in-place and provides a high-level layer of management abstraction for them.

Mesosphere

Matt Trifiro, senior vice president of Mesosophere, explained in a phone call the key differences between the enterprise and community versions of DCOS. The enterprise version "runs anywhere," he said -- on bare metal, in containers, on top of OpenStack, or in the cloud provider of one's choice.

The free-to-use community version is currently deployed only as an AWS image (Google Cloud Platform and Azure to follow), but will eventually be released as an asset that can be used in one's own data center, although the timing has not yet been decided.

Right now, the emphasis is on deploying to public clouds -- the venue where DCOS's feature set will likely pay off in the biggest way.

Bigger bangs for the buck

The for-pay enterprise edition of DCOS, as the name implies, includes features specifically aimed at enterprises: Kerberos-based authentication, SLA support, and support for hybrid cloud environments.

But where DCOS's enterprise edition will really shine, said Trifiro, is through a series of future projects to solve problems specific to enterprise data centers, wherever they live. One such feature is oversubscription, which expands on DCOS's existing ability to boost average utilization in a data center.

Said utilization is typically at around "eight to 15 percent," he said, "so about 85 percent of your data center goes wasted. Just by installing DCOS, you can maybe double or triple that [utilization], because of more efficient scheduling and running multiple workloads on the same machines."

With oversubscription, a technology derived from work done at Stanford University, "organizations will be able to get easily upward of 70 or 75 percent," Trifiro said. He described how one of DCOS's users, running on thousands on nodes, was able to wring 95 percent utilization out of that cluster. (Another customer, running on AWS, was said to cut its bill by 65 percent.)

Provisioning for the public cloud

The scheduling and provisioning system in DCOS, Trifiro explained, has separate schedulers for each of the applications and services in the cluster. "They're making decisions about what kinds of resources they need at any given time," he said, "based on their own algorithms." DCOS, in turn, arbitrates between resource requests for all the different running apps and can even be addressed programmatically by a given app if it's DCOS-aware.

Features like this underscore how DCOS's public-cloud pitch is meant to pay off. Public cloud use with enterprises is growing at a striking pace thanks to the flexibility of infrastructure it provides. DCOS's way of wringing the most usage out of the least number of machines, without changing the underlying OS, complements those who already have an existing public cloud investment and want to keep an eye on costs, even as prices there spiral downward.