Univa today announced general availability of Grid Engine 8.4.0. The latest version of Grid Engine includes many new features including expanded support for Docker containers as well as “preview support” for Intel’s latest Xeon Phi code named Knights Landing processor. Univa also reports fixing more than 80 prior issues.
Leading the container enhancements, users can now automatically dispatch and run jobs in Docker containers, from a user specified Docker image, on a Univa Grid Engine cluster. Univa says this simplifies running complex applications in a Grid Engine cluster and reduces configuration and OS issues. Grid Engine 8.4.0 isolates user applications into their own container, avoiding conflict with other jobs on the system, and enables legacy applications in Docker containers and non-container applications to run in the same cluster.
“Adding Docker support to Univa Grid Engine allows our customers to easily submit a job and create a container with one simple Univa Grid Engine command. Users can create containers from images on the public Docker Registry or configure their own private registry with their own images,” said Bill Bryce, vice president of products at Univa.
In terms of developing early support for the latest Intel Phi processor, Univa – like many others – didn’t receive a KNL development machine from Intel until quite recently. “This is our initial implementation and it is similar to what we did for the previous Xeon Phi Xeon,” said Bryce.
The early KNL support allows Grid Engine users to obtain various metrics and data from KNL and to use that when scheduling work onto the new Xeon Phi. Univa Grid Engine can launch and control jobs on Intel’s KNL machines, extending the current Intel Xeon Phi integration to the latest KNL processors. This update simplifies running and managing Intel Xeon Phi based applications on the cluster according to Univa.
“We do expect there will be more work to do in terms of how to better manage the memory on KNL than we are doing right now,” Bryce said. That will wait until the next round of more stable silicon is available. Currently, developers are still in the early stages of gaining wider access to KNL. (See HPCwire article, Intel Debuts ‘Knights Landing’ Ninja Developer Platform.) As part of Intel’s steady rollout of KNL related material, the company just announced an updated text for KNL programming, Intel Xeon Phi Processor High Performance Programming: Knights Landing Edition.
Additional Docker support improvements to Univa Grid Engine includes:
- Docker Directory Mapping – maps user directories outside of a Docker container into a Docker container during a job run, making it easier to transfer files
- Full Job Control of Docker Containers – provides Grid Engine with the ability to suspend, resume or kill existing jobs running in containers
- Full Job Accounting for Docker Containers – provides accounting information for the job running in the Docker container
- Run Parallel Jobs in Docker Containers – Users can submit parallel jobs to Grid Engine that will run on multiple machines in Docker containers
- Automatic Handling – input, output and error files for Docker jobs
- Run Interactive Applications in Docker Containers – Grid Engine Interactive sessions can run inside a container
Another key upgrade, says Univa, is that administrators can examine different aspects of Univa Grid Engine 8.4.0 to determine if a type of job, configuration or infrastructure outside of Univa Grid Engine is causing issues and slowdowns with the system profiling feature. This reduces the time spent diagnosing issues in clusters and provides a much needed tool for Univa Grid Engine Administrators.
“Grid Engine 8.4.0 is a significant upgrade for us with the support and integration of Docker containers,” said Bryce, “We recognize the need for developers to manage all of the different moving parts of an app and funnel in their updates to reduce complexity and run more efficiently.”
The use of container technology has been steadily growing in HPC, including recent development of Docker alternatives such as Singularity at Lawrence Berkley National Laboratory (see HPCwire article, Custom App Container Targets HPC Needs) as well as development of Shifter at NERSC (see HPCwire article, NERSC’s ‘Shifter’ Makes Container-based HPC a Breeze).