In the early days of computing, the goal was to automate and speedup the job being done by human calculators. Early computers were large and complex devices, for the period, and programmers and operators required an intimate knowledge of the workings of these machines to achieve the desired results. Indeed, in works that describe the early days of computing such as Turing’s Cathedral: The Origins of the Digital Universe by George Dyson, we get an intimate look at and appreciation for the computing skills those early pioneers had – in addition to their domain specific knowledge. Dyson writes: “The ENIAC was programmed by setting banks of 10-position switches and connecting thousands of cables by hand. Hours, sometimes days, were required to execute a programming change” (75).
Fast forward to today, and supercomputers or high-performance computing (HPC) clusters have evolved to have massive computational abilities – and this power is applied to a mind-boggling array of challenges across numerous industries, with an increasing focus on artificial intelligence (AI). As user communities of these environments grow, so do the challenges in bringing to bare these crucial resources to best serve the needs of an organization.
HPC environments today are rapidly growing in scale and complexity. Today it’s common to find high-end computing environments containing thousands of nodes, with hundreds of users running millions of jobs per day. Computing environments are frequently heterogeneous in nature, containing a mix of processors, accelerators, interconnect, and storage types all with the goal of delivering performance. From this complexity, three primary concerns arise – usability, efficiency, and oversight.
The User Experience
As high-performance computing has evolved, so have the people that make use of it. Scientists and engineers today are highly focused on research and discovery and are not experts on computer hardware like their predecessors were. For them, HPC is a means to delivering value to the business. From punch cards to keyboards, mice to touch screens, there has been a significant evolution of technology user interfaces over the decades. Today, we are the swipe, pinch and drag generation. Devices ranging from mobile phone, to tablets to personal computers all make it easier than ever to interact with systems. Why should users of HPC have it any different? Indeed, organizations today with an investment in HPC, need to take into consideration not only raw FLOPS, but also the usability of the environment. Users need to be able to submit & manage their work simply, and reliably – regardless of where they are.
Walking the Tightrope
Getting access to resources in a timely fashion is often a tug-a-war in HPC environments. Aligning business needs with demands from the varying projects, groups, users can be complex to say the least. The competitive nature of business today precludes relying on first come, first served for running your HPC jobs. Overall the infrastructure needs to be kept at a boil, while keeping your user community cool. Like the proverbial traffic cop, HPC environments require intelligence to make workflows flow, push projects, eliminate workload and data traffic jams – all to keep the business happy.
Demystifying Your Environment
We’ve discussed the inherent complexity in HPC environments and the importance of not only using the resources in an intelligent manner that’s well aligned to the business, but also making those resources as easily accessible as possible to your user community. Now, let’s shift our focus to the heroes responsible for keeping the environments in top running condition – the administrators. Administrators are the ones we call when jobs fail – which can be for any number of reasons from a hardware fault, insufficient software licenses, to a full filesystem. Ideally, HPC environments require tools for insight which tie together both infrastructure, job, and software license details for a comprehensive view. This can enable administrators to quickly get users back up and running when time counts.
Certainly, one could opt for a collection of different tools to fulfill the above requirements. But this often requires additional work around integration, and results in having several different support providers when things go wrong. Given these considerations, a holistic software solution for managing a high-performance computing cluster can provide significant benefits.
IBM Spectrum LSF Suites is a tightly-integrated, systems and HPC workload management solution which provides leading scheduling efficiency and a simplified user experience for administrators and users alike – all rolled into one. Built upon 25 years of expertise in workload scheduling and backed by IBM, Spectrum LSF Suites grow with your business needs – with 3 available editions for progressively larger sites with an increasing level of capabilities. Get piece of mind in a hectic HPC world with IBM Spectrum LSF Suites.