Altair HPC

Thoughts on High Performance Computing (HPC) from the Altair PBS Works team -- our vision of the future, our favorite tips and tricks for getting the most out of Altair's products, and our take on the latest HPC trends (Cloud, Green, GPUs, Exascale, Big Data, ...).

Stay up to date, follow our HPC RSS Feed

Content Background Top
Category Archives: HPC




1 Comment
avatar Category: HPC

5 Predictions for HPC from the 2013 PBS Works User Group

, CTO, PBS Works division at Altair

For me, what makes the PBS Works User Group fun is the really great group of people that come together across a broad spectrum of the HPC community.  I thought this was an excellent forum in which to get real insights from the real community.  So, this year, we closed the event with a general customer feedback session, during which, I asked:  Where is HPC headed in the next 5 Years?

In short, there was consensus in 5 areas — from the real experts:

1. Most end users will not be supercomputing savvy:  the number of end users will double in the next 5 years, but most of the new users will be new to HPC and even new to Linux (e.g., when you mention MPI or ssh, they won’t know what you’re talking about).

2. Capacity will triple and power constraints will drive GPU adoption:  more cores, more memory, more bandwidth, …, and a lot more systems will have 1 or 2 or 3 GPUs due to power constraints.

3. Big organizations will consolidate HPC centers, making clusters bigger and driving adoption of remote visualization:  multiple organizations have plans to consolidate HPC sites spread around the world into a handful of larger sites, also merging smaller clusters into larger clusters.  Six of the larger organizations said this means they will heavily depend on Remote Visualization.

4. Microsoft Windows use will increase in HPC, entirely driven by software:  among the larger organizations who must support a wide variety of ISV applications, Microsoft Windows support is required, as certain ISV applications only work on Windows

5. Adoption of public clouds for HPC is still an open question:  despite “expert” pronouncements that public HPC clouds are the future (and I would sheepishly include myself in this category), the general consensus of the actual HPC community is that, even 5 years from now, less than 10% of HPC cycles will be delivered via public clouds.

I was both surprised and gratified that 4 out of the 5 predictions were nicely lined up with the vision we have been pursuing at Altair for HPC.  As for the 5th, we are investing in public HPC Clouds too :-) .  If you are attending the SC13 conference in Denver, CO later this month (, please stop by the Altair booth and say “Hi” — you can also get a demo of how Altair is addressing the future of HPC.





If you’ve been keeping up with the latest trends in PBS Works, you’ve probably heard of the PBS Application Services (PAS). PAS is a piece of software that sits on top of your PBS Professional complex, exposing a collection of web services that allow front-end products such as Compute Manager and PBS Desktop to submit, monitor and manage jobs on your remote PBS Professional complex. In addition to all of these cool web services is an extremely powerful feature that PAS provides called Application Definitions (app-defs). App-defs are, put simply, a well structured mark-up language for describing your application (i.e. OptiStruct, RADIOSS, AcuSolve, LS-Dyna) to both PBS Professional and front-end products wanting to provide a cohesive interface to submitting users.

Although Compute Manager, PBS Desktop and PAS give rise to an amazing leap forward in end-user job submission, these technologies *could* add an additional maintenance burden to PBS Professional administrators. One major culprit here is the fact that now there are two or possibly even three entry points to your PBS Professional complex, and potentially two or three different variants of the same application (e.g. let’s say OptiStruct in this case). Many organizations do not want to remove the well proven and established command-line interface to their complexes, yet still want to take advantage of the more cutting edge approach of graphical submission, specifically for the younger generation of engineers and scientists. A common mind set might be, “I’ll leave my existing OptiStruct job scripts for the command-line, and author a new OptiStruct app-def” for say, Compute Manager.

Here’s a short list of possible issues people face with this mind set…

1.) In many cases the OptiStruct application available through Compute Manager is not consistent with the OptiStruct job script people are using via the command-line.

2.) Jobs submitted via the command-line don’t display properly via the Compute Manager monitoring page.

3.) Job files available in the job execution directory are not exposed to Compute Manager, essentially removing any remote file browsing functionality while the job is running.

It’s very important to keep in mind that all of the above issues in this approach could be solved with additional effort and changes to both variants of OptiStruct. And there may be some site-specifc requirements which encourage you to maintain two OptiStruct applications scripts/app-defs. But why go through all of this effort to maintain two OptiStruct applications and manage how they interoperate with one another if you don’t need to? The first step in walking away from this mindset is re-visiting a few notions that you may already have in your mind, and that is; PAS and app-defs aren’t just for graphical front-end products, they can be leveraged via the command-line interfaces as well.

The PAS team recently introduced a set of command-line tools that bridge the gap between GUI and CLI, opening up an entirely new way of approaching the authoring and maintenance of applications ran on PBS Professional complexes. Put simply, the very same OptiStruct app-def that is used in your installation of Compute Manager, can also be used by your command-line users who are more comfortable submitting their simulations via the command-line, removing the need for submitting via ‘qsub’ entirely. This is achieved by submitting an OptiStruct job using the ‘pas-submit’ command. Now submitting engineers and scientists have a common experience to submitting OptiStruct, regardless of which entry point to PBS Professional they are using.







If you ever used a workload manager to schedule commercial software runs on a computing environment, you almost certainly faced the problem of software license competition: you really need to only let a job run when there are available licenses.

Many commercial job schedulers offer a “dynamic resource” mechanism to handle such cases and PBS Professional does as well.  But, in addition, PBS Professional offers the ability to use a plug-in to handle corner cases and complex license requirements, not easily expressed in terms of resources.

A PBS Professional plug-in is a script associated to an event, and in the case of license scheduling we can leverage the “runjob” event: this happens when a job is found able to run on the available resources, just before it is effectively dispatched. “Runjob” is a good time to check for certain health status information on the nodes selected for running the job, to check for accounting information (has the user enough allocation units left?), and in our case, to check for license availability.

The flow would be:

1. Job is submitted asking for a certain amount of “scheduling resources” (ie. ncpus, memory) and holds also the necessary information to express the license needs (ie. license server pointer, needed amount of features) but license related resources are not used by the scheduler.

2. When scheduler finds there are enough scheduling resources available, job is dispatched.

3. Just before dispatch, the “runjob” event is evaluated: as the plug-in is able to look into job data, it reads the licensing information, connects to the license server and lets the job be dispatched only if enough licenses are available (taking care to also minimize race conditions, for instance throttling the amount of jobs dispatched per minute).

Of course this is not perfect (as the dynamic resource mechanism, anyway), but I found I could handle tricky commercial software license requests without placing load in the scheduling mechanism (for instance license related information is processed just when needed and not at each cycle) or using complex resource configurations.

A real world example of the above comes with the document describing how to integrate Schlumberger Simulators in a PBS Professional Computing Environment (Altair-PBS-Eclipse-Integraton-2012.pdf), which is available from (Support -> Resource Library -> Technical Papers).





This year Altair held it’s first annual PBS Works User Group October 2 & 3, 2013, in the Tech Museum of Innovation in San Jose, CA, a great background for the event. During breaks from discussing the next generation of HPC computing attendees were able to wander around and see some of the cool exhibits showing technical milestones of the past. One of the things I was surprised by was the effect of having all the school aged children there in the museum while the conference was going on. Rather than being a distraction, their curiosity and energy helped get me get excited about what the future would bring.

The presenters at conference were drawn from a wide range of backgrounds, from academic to industry to government. Martin Nichols came all the way from the University of Queensland in Australia to talk about how he’s using PBS Professional hooks to make sure their various departments and groups have access to the compute resources they need. Simon Burbidge from the Imperial College of London made the long trip to talk about his vision of where HPC was going. Jim Glidewell from Boeing Corp from up the coast in Seattle, Washington, was there to talk about his 12 years of experience using PBS Pro on their ever changing HPC infrastructure. Jun Iyama came from Nissan Motor Co in Japan to tell us about how PBS Pro was being used in their engineering environment to optimize running their incredibly complex crash simulations. Dale Talcott and Greg Matthews came down the street from NASA-Ames to talk about how their rocket scientists are using PBS Professional on one of the largest computer systems in the world. And many more incredibly talented individuals from all over the world, it really was a global event. Presentations from the conference can be found at

In addition to the great presentations there were also several opportunities for attendees to participate in highly interactive sessions with Altair engineers and developers. A core group of PBS Professional developers held a Q&A discussion with the entire audience. Separate optional tracks were offered for advanced PBS Pro hook writers as well as a workshop for creating application definitions (AppDefs) for Compute Manager, Altair’s cool new graphical front end to PBS Pro. It was a great opportunity for those using our technology to learn something new and for Altair to hear what improvements could make it easier to implement.

Finally I’d like to say thank you to Rich Bruekner of insideHPC for being there live and allowing us the share the proceedings real time. He’s got some great additional interviews and articles on their site: PBS on inside HPC

Next years PBSWorks UG will be held in conjunction with the Altair ATC in Irvine, CA, on Oct 1-3 2013, hope to see you there!

Sam Goosen




Category: HPC

Welcome to the New Altair HPC Blog

, CTO, PBS Works division at Altair

It gives me great pleasure to re-start up the Altair HPC blog.  Didn’t know we had a blog dedicated to the HPC space before?  I don’t blame you — we’ve mostly posted product announcements — not really very blog-y.  The PBS Works product management team hopes to change that, and provide a place where you can get a more personal view from us.  Hear our view of HPC in general, see our vision of HPC products and solutions, learn  our favorite tips and tricks for using our software, but most importantly, simply get to know us a little better.


Before getting started, let me give you a little insight into the PBS Works product management team:

Bill Nitzberg (me) — easily recognizable at conferences by his suspenders and aloha shirt…

Dan Stephenson — jumped out of airplanes in his previous line of work…

Sam Goosen — has been known to go skiing from time to time…

Dario Dorella — recently discovered that jetlag and marathons don’t mix…


Please let us know what you think of the new format… although you might wait for a few more posts :-) .



Content Background Top

© Copyright 2012 Altair Engineering, Inc. All Rights Reserved.