Efficient software for more efficient computing

What exactly does KONWIHR support, and who does it support?
Bungartz: “Anyone conducting research at a Bavarian university can apply for funding. KONWIHR provides funding and expertise for work on scientific codes. Depending on the issue at hand, there are two funding formats: Up to €10,000 for three months, or up to €50,000 for one year. Each project is also assigned a contact person at one of the computing centres. KONWIHR aims to strengthen expertise and the Bavarian HPC network. The goal is to establish a peer-to-peer consulting system in which specialists at the computing centres connect researchers with scientists who have previously solved similar problems with the help of KONWIHR and can now assist in accelerating tasks.”
Wellein: “Through KONWIHR, project staff can work directly at one of the two data centres for a period of time and solve software- or implementation-related tasks in direct contact with specialists.' This has unfortunately not been used as often since the start of the pandemic, but research groups using high-performance computing for the first time benefit greatly from it. In addition, two KONWIHR workshops are held each year, where applicants present their projects to a larger audience. These workshops also serve to promote networking and knowledge transfer.”

What issues does KONWIHR face?
Wellein: “One major issue is certainly porting code to a different system or computer architecture. Another is code optimisation: software should be more efficient and run on more parallel nodes, or the work of all the computing nodes used should be accelerated and improved.”
Bungartz: “KONWIHR also promotes functional enhancements to software. For example, climate simulations may be supplemented with additional parameters or analysis options may be incorporated into a simulation code. If two codes need to be linked for a simulation and an elegant interface is required, that is a project for KONWIHR.”
Wellein: “The team may also have a workflow that runs on Windows laptops and now needs to be executed on a high-performance computer. KONWIHR supports clearly defined research projects that require technical solutions.”

What constitutes a classic KONWIHR project?
Bungartz: "The prerequisite is always a scientific question that can only be reasonably answered with the help of a computer. KONWIHR focuses on the technical side, so there must be potential to optimise the code."
Wellein: “All applications are reviewed by two specialists. One looks at the scientific question. The other evaluates the use of HPC: is the technology or code suitable? Does a higher resolution really improve the calculation? Is the work plan coherent? It is entirely possible for the two specialists to give diametrically opposed assessments: For example, the subject matter expert may support the project because the task is new, while the computer specialist may criticise the fact that the simulation code is not at all parallelised and therefore cannot be executed.”

KONWIHR's goal was to promote the use of HPC in research and science. Has this task not already been more than fulfilled? 
Bungartz: „You could see it that way. Twenty-five years ago, only a few experts modelled gas flows in the universe or complex technical processes. Today, simulations are standard tools in the natural sciences, life sciences, and engineering. However, technology continues to evolve. Data analytics and artificial intelligence methods are currently bringing high-performance computing to the social sciences and humanities. Typically, it is high computing power that is in demand, but this is referred to as ,high-performance computing’ rather than ,high-performance computers’. The focus is not on using the largest computer systems, but on getting the most out of them.”
Wellein: “In addition, many researchers inevitably end up using high-performance computing. Most of them initially programme solutions for their data queries on their laptops. If these work well, the technical resources eventually become insufficient because it takes weeks to get results. This is where KONWIHR and the computer centres step in: we provide the team with a larger system so they can perform more complex calculations.”
Bungartz: “KONWIHR also helps to overcome inhibitions. Through the programme, researchers at a university in a rural area realise that the high-performance computers at the LRZ or the NHR@FAU are also available to them. Last but not least, students and doctoral candidates graduate, new professors are appointed and new people arrive who need practical HPC experience to answer their research questions.”

Which developments have shaped KONWIHR over the last 25 years?
Wellein: “The parallelisation of computing nodes with the Message Passing Interface, or MPI, still has an impact today. Then, the demand for support in the area of numerical simulation and algorithms increased significantly. Now, the integration of accelerators and their impact on codes is a major topic. Over the past year or two, we have also received requests for support relating to the interaction between AI and HPC, as well as the use of AI models.”
Bungartz: “KONWIHR's goal of achieving more efficient computing has been embraced by the simulation-oriented HPC community, resulting in a wealth of knowledge in this area. The AI world is only just beginning to consider efficiency, so in that sense, we are currently experiencing déjà vu. Twenty to twenty-five years ago, we already had researchers who wanted more and more CPU or computing power. Now, the 'give me more' demand is increasingly coming from the AI community. As we gain more experience, it is becoming clear that performance increases are driven equally by hardware, numerical linear algebra, and algorithms.”

Do the higher funding amounts go to the LRZ because that's where the larger, more complex computers are located?
Wellein: “The grants have nothing to do with the size of the systems, but rather with how long researchers work with a particular piece of software. It is often observed that 3-month projects are used to hire a student assistant for adjustments. For larger applications, the necessary software and knowledge are already in place. These are often projects supported by the German Research Foundation for three or four years, for which someone is needed to take care of the software or technology. With €50,000, you can employ someone for around eight or nine months and ensure that the project produces scientific results and achieves sustainability in terms of software.”

How is KONWIHR responding to accelerator technologies, whether with GPUs, quantum systems or new analogue technologies?
Wellein: “These developments are primarily reflected in the applications.' We are now seeing the first exploratory projects on quantum computing. The technology is changing, but the goal remains the same: to make applications, code and workflows more efficient, faster and more economical. The situation with AI methods and data analytics is comparable to HPC 30 years ago. The difference is that, back then, everyone had their own code. Today, standard frameworks are widely used for AI. However, as soon as methods or codes need to be differentiated and adapted, we can draw on our expertise built up over the last 30 years.”

In addition to new technologies, energy efficiency is also coming into focus. Is this a topic covered by KONWIHR applications?
Bungartz: “KONWIHR is and always has been a tool for ensuring sustainability and energy efficiency. If you can solve a scientific problem in ten per cent of the usual time, you also save 90 per cent of the energy. Research groups don't submit applications for energy efficiency per se, but rather to improve their applications.”