The Easy HPC button
We live in a results-driven world. Whether it’s an aerodynamicist waiting on simulation results to determine the efficiency of their latest model, or a doctor waiting on genomic pipeline results to determine next steps for a patient, results make the world go round. And this of course goes beyond the sciences. As any thespian will tell you, stage productions are the result of the work of many individuals behind the scenes.
Much in the same way, complex computational processes that are found in HPC rely upon many things behind the scenes to be carried out. And although the devil may be in the details, consumers of HPC resources shouldn’t have to go through purgatory to get results. Organizations today rely on HPC to drive their core mission, delivering products to market faster. So, it goes without saying that the need for HPC to be easy to drive productivity is crucial. And much like the technology of HPC has changed so have the skills of the users. Modern HPC infrastructure relies upon a myriad of technologies including containerization, accelerators and cloud. And for users, gone are the expectations of learning a complex CLI, replaced by the need for easy-to-use interfaces.
Workload schedulers are a necessary component of any HPC cluster. Schedulers have been around for a very long time and as they become more sophisticated, they support an ever-increasing number of CLI and configuration options. Although these options provide greater functionality, their use can be complicated to end users. What if you could provide an HPC easy button for your users?
IBM Spectrum LSF is a workload management solution for HPC environments. Over the past 30 years, it’s evolved from being just a workload scheduler, to an entire suite of capabilities covering the lifecycle of HPC jobs. Scheduling wise, LSF has not only kept pace with the massive scale of commercial HPC environments today, but also provides capabilities which dramatically lower the bar to access HPC.
Ease of use starts with the users and LSF provides a web-based job submission and management portal which greatly simplifies the use of your HPC cluster. Administrators define custom forms that hide the complexity, and they can even be customized to use application and domain specific language understood by your users. For users on the go, LSF has Android and iOS mobile clients so you can check on the state of your running jobs. And a RESTful API is also available to integrate LSF into your corporate infrastructure.
With users well taken care of, LSF features many capabilities which allow administrators to take advantage of technologies such as containerization, hybrid cloud and GPUs. Out of the box support for various container technologies let’s administrators control which containers can be used in the environment and hides the complex container startup commands from users. Support for dynamic hybrid cloud enables LSF to burst out to any of the supported cloud providers when needed and scale back the resources when no longer required. And intelligent data staging takes care of moving data to and from the cloud without blocking or making resources wait for transfers.
What does this all add up to? Well, you can think of it as an HPC easy button. Your users simply fill in a form and submit their job. LSF worries about the underlying complexities, where to place the job, moving data, CPU and GPU allocation. The user waits to get the job results back and is oblivious everything that is going on behind the curtain.
Learn more about easy HPC with IBM Spectrum LSF in this session: Simplifying HPC - Just push the button.