IBM Builds A Better OpenStack Scheduler

IBM has launched a proprietary product to work with OpenStack's resource scheduler, the part of the cloud software stack that assigns workloads to virtual machine hosts in the cloud.

IBM has a history of mapping out resource schedulers, thanks to its experience in mainframes and with Tivoli Workload Scheduler LoadLeveler. In addition, the company acquired 20-year-old Platform Computing two years ago, adding its Load Sharing Facility to IBM's scheduler product line.

In making that acquisition, IBM, a supporter of the OpenStack open source code product, is saying unequivocally that OpenStack's scheduler remains a rudimentary model of the breed and users will need to buy into IBM's expertise to get the most out of their OpenStack clouds.

OpenStack's scheduler is static, looking at information in the Nova SQL database before assigning a workload to a server host. (Nova is the provisioning and scheduling component of OpenStack.) But it looks only once, when in fact the amount of work that applications are doing on the host is constantly changing. That leads to what Gord Sissons, senior manager of technical marketing manager and a Platform veteran, calls OpenStack's "Goldilocks problem." Some hosts running highly active applications become "too hot," while others with a similar number are doing much less work and are "too cold."

[Want to learn more about why OpenStack might be chosen for private clouds? See VMware, Mirantis Put OpenStack On vSphere.]

IBM's Platform Resource Scheduler works on top of OpenStack's to manage workloads in a more dynamic fashion, Sissons explains. "Platform has more of a real-time view. It knows about a whole set of scheduling metrics that Nova [doesn't know about and] ignores."

Platform can use parallel assignment techniques to distribute workloads over a cluster. When it detects an application is resident on a host with too few resources, it moves the virtual machine to another, less utilized host. It uses VMware's vMotion, KVM's Live Motion, or Hyper-V's Live Migration function to move the running virtual machine.

Once the OpenStack scheduler makes a workload assignment, its view of the job is static -- it isn't looking for feedback on how hard the server is working. "It doesn't think about rescheduling," Sissons explains.

Platform Resource Scheduler became available in technical preview in November as part of IBM SmartCloud Orchestrator, IBM's cloud provisioning software. It's now available as a separate product that serves as an optional add-on to Orchestrator or IBM SmartCloud Entry, both of which are used in managing OpenStack clouds. IBM charges for the scheduler based on how many cores it's running on, but the company declined to offer a per-core price with the announcement. Pricing details depend in part on the customer's implementation, according to Sissons.

Charles Babcock is an editor-at-large for InformationWeek, having joined the publication in 2003. He is the former editor-in-chief of Digital News, former software editor of Computerworld and former technology editor of Interactive Week.

Private clouds are moving rapidly from concept to production. But some fears about expertise and integration still linger. Also in the Private Clouds Step Up issue of InformationWeek: The public cloud and the steam engine have more in common than you might think. (Free registration required.)