Two big announcements: the open source project name is changing to “OpenPBS” and version 20.0 is coming in the next few weeks, and includes a great set of new features and fixes.
The PBS software is dual-licensed — a single core technology, available under both an open source license and a commercial license — and this approach has been truly successful, with innovations flowing between researchers and enterprises, and with a large and growing community, globally. We continue to be 100% committed to this strategy — the core job scheduling and workload management technology will continue to be “a single core” and dual-licensed. However, to reduce “version confusion”, we are giving each version a distinct name: OpenPBS will be the open source licensed version, and PBS Professional (or PBS Pro) will be the commercially licensed version.
Over the next few weeks, we will be making changes to update the name to OpenPBS, including moving the main project page to openpbs.org, moving the source code repos to github.com/openpbs, updating the RPM names, and a lot of other stuff. The changes will be made in as seamless a way as possible, but, as with any major update, you may encounter an issue or two; we ask that you be patient and report any issues on the community forum.
The first release under the OpenPBS name will be version 20, and we are targeting it for late April / May. And, wow, it’s another big release, including:
- Faster scheduling via multithreading, better job dispatch (~30x faster), and better preemption (~2x faster)
- Faster/better turnaround by allowing a job which could run on multiple different combinations of resources, start on the first/best available (via the Beta “runone" dependency)
- Faster, more scalable, hook throughput via pre-cached job data and a non-blocking execjob_end event; plus, additional hook events, including pre-resume and post-suspend
- Native PBS support for the Cray Shasta platform
- Numerous advance reservations enhancements:
- Admins can reserve nodes for maintenance via “force” creating “maintenance” reservations
- Long-waiting jobs can “keep their place in the queue” despite startup faults or human errors by converting these jobs to new “start reservations” (which an admin can do automatically)
- “Idle” reservations can be automatically ended to free resources
- Reservation owners can add users/groups and, when allowed, change durations
- Support for MultiSched and replacing “down” nodes in running reservations
- Accounting records now include full submission information (including dependencies), updates on qalter, and suspend/resume event information
- Plus, improvements for cgroups, job arrays, qstat, preemption, MOM footprint, and more. View the full list of enhancements and bug fixes via the commit logs.
As we move forward with the transition, and with the release of version 20.0, I’ll post updates to the community forum, so “watch this space”!
On a personal note (during this crazy and troubling C-19 time), please be safe, wash your hands, and take care of yourselves and your families. And, to those of you on the front lines — THANK YOU!
Bill Nitzberg, OpenPBS Community Manager