Analysis / News / /

Univa Brings Supercomputer Scheduling to Kubernetes

31 May 2016 5:08am, by

As the market about container scheduling heats up, Univa, workload management specialists from the high-performance computing (HPC) market, are among those entering the ring.

Earlier this month, the company demonstrated Navops Command, a pluggable, automated workload placement and policy management solution for Kubernetes, at OSCON conference in Austin, Texas

“Navops Command’s big differentiator is its rich policy management which allows a Kubernetes admin to have a lot more control over what runs where when,” said Robert Lalonde, vice president and general manager of Navops. NavOps Command offers Kubernetes-enhancing features access control, fair share resource placement, workload prioritization, pre-emption, and a variety of placement policies provide for enterprise-grade scheduling.

Univa launched its Navops container-management suite last year, based on the Grid Engine cluster workload scheduling technology originally created at Sun Microsystems and later acquired from Oracle. The company realized that the Grid Engine core technology around provisioning and managing workloads in HPC was going to be relevant in the container world as people started building microservices. The technology can scale compute resources across on-premise, hybrid and cloud infrastructures.

The company also supports Docker with its Univa Grid Engine Container Edition. Its Navops Launch offers a quick-start package with enterprise-grade management controls for Docker containers.

command-after

Navops Command enables companies to use any Kubernetes-based system, such as RedHat’s OpenShift and CoreOS’s Tectonic. In a blog post, CoreOS’s Hongchao Deng recently discussed its efforts to improve upon the Kubernetes scheduler.

Kubernetes offers support for multiple schedulers, giving customers the option to use the one that suits them best, explains Lalonde. The Kubernetes scheduler is turned off, and Navops Command plugged in instead. It competes with the likes of Hashicorp’s Nomad and Google Omega, which are designed specifically for containers, and IBM’s Platform LSF scheduler and Adaptive Computing’s Moab, which are HPC solutions that have been adapted to containers.

“We have millions of cores among our customers. Some are running millions of jobs; some are running thousands,” Lalonde said. “Every single customer wants to do it a different way. They have different rules about how they want to manage that workload, how they want to place it, track it, how they want to bill their clients and internal clients for using the clusters.

Refining Kubernetes

Navops Command’s capabilities include advanced scheduling algorithms, sophisticated policies for managing SLAs and the ability to prioritize priorities when resources become scarce.

“Microservice applications decompose into many, many containers and those could be 50 or 1,000 containers. Each one of those containers could have multiple instances,” Lalonde said. “You might be starting up shopping cart containers vs. tax modules. There would be any number of things you’re running on your website, and the management of that becomes very complex.

“You need to make sure things are running at certain ratios; you need to make sure certain things are running at certain times of day. As the sun moves around the globe, that workload’s coming from different countries, and that implies different user interfaces being used, potentially different languages, certain tax modules being active at certain times of day. That’s where advanced policy management lets you set rules about what’s running when.”

Univa was a part of the conference dialogue about schedulers at KubeCon 2015. It was a founding member of the Cloud Native Computing Foundation, which aims to harmonize emerging technologies for cloud applications and operations.

Navops Launch includes a subset of the company’s UniCloud provisioning tool, Puppet system configuration tools, Docker containers, Kubernetes container and pod management and the minimalist Fedora development release of Atomic Host Linux from Red Hat.

Its Grid Engine customers include Panasonic, United Airlines and Siemens, though it’s too early to name Navops customers, Lalonde said.

However, it’s touting research by the Samsung SDS America Cloud Native Computing Team comparing Navops Launch to provisioning a bare-metal Kubernetes cluster by hand. That test found installation with Launch more than five times faster for large clusters and that it eliminated two-thirds of the required steps.

CoreOS, Docker, IBM and Red Hat OpenShift are sponsors of The New Stack.

Feature Image: “Rules” by Marcin Wichary, licensed under CC BY-SA 2.0.


A digest of the week’s most important stories & analyses.

View / Add Comments