Jonas Lindemann discusses how remote access to HPC resources helps make researchers' lives easier and enables new users to access HPC more easily.
LUNARC, the centre for scientific and technical computing at Lund University in Sweden, provides the university and the wider Swedish research community with access to advanced HPC resources.
These computing systems are designed to enable large-scale scientific simulations, complex data analysis, and high-end visualisation tasks that would be impossible to perform on ordinary desktop computers or workstations. By operating powerful clusters equipped with both CPUs and GPUs, LUNARC supports a wide range of research disciplines and helps drive forward computationally intensive science.
The centre plays a central role in supporting research at Lund University. It offers researchers not only access to hardware but also an integrated computing environment that combines active storage, interactive tools, and a queuing system for running large-scale jobs.
One of the distinguishing features of LUNARC is its emphasis on accessibility through interactive environments. From the beginning, users could log in through the command line to compile and run software. Over time, however, the centre recognised the need for a more user-friendly approach. It developed a desktop environment that allows researchers to access the systems as if they were working on their own personal computers, but with the benefit of vastly greater computational power and storage. This environment supports familiar tools such as Visual Studio Code and Qt Creator, as well as file explorers and graphical interfaces for visualisation. Importantly, it also provides access to hardware-accelerated graphics, making it possible to perform advanced visualisation tasks without the need for a powerful workstation at home or in the laboratory.
The remote desktop has changed the way researchers interact with high-performance computing. Instead of constantly transferring data between local machines and the cluster, users can keep their data on LUNARC’s systems and work directly with it. This reduces bottlenecks associated with data movement and ensures that large data sets, sometimes reaching hundreds of terabytes, can be managed efficiently. Researchers can carry out their work from any location, whether in the office or at home, simply by downloading a client and logging in. In recent years, the adoption of this desktop model has grown steadily, and a significant proportion of users now rely on it, particularly those running visualisation software such as ParaView or commercial engineering tools.
Looking to the future, LUNARC is also exploring web-based portal access, which would further lower the barrier for new users and provide an environment especially well-suited to artificial intelligence and interactive workloads. This builds on a consistent philosophy at the centre: to combine the power of traditional HPC systems with user interfaces and workflows that researchers find approachable and productive.
Can you tell me about Lunarc and your role at the facility?
Lindemann: Lunarc is a centre for scientific technical computing at Lund University. Our role is to provide computational resources with active storage, rather than long-term storage. In addition, we provide an environment with a queuing system for submitting jobs as well as facilities for interactive use. Our desktop system serves as the main entry point for users.
We have around 1,000 active users. The majority work in chemistry and molecular simulations, about 20 percent are in engineering, and the rest represent a broad mix of disciplines. Recently, we have seen a strong increase in users coming for GPU and AI applications.
Has the proliferation of GPUs changed how you manage resources?
Lindemann: GPUs and AI have broadened the appeal of HPC resources to users who may not have traditional HPC backgrounds. Some of these users are very advanced, often coming from cloud computing environments, and they adapt quickly. Others have simply outgrown their laptops and workstations and need more powerful resources. They tend to be less familiar with HPC systems, which means we provide additional training to help them get started.
We identify training needs in several ways. Much of the input comes from our application experts, who support users directly and see where challenges arise. We then tailor general courses and documentation accordingly. We also collaborate with the Graduate School for Numerical Computing, where many of our users are based. Together, we run programming courses, mainly in Python, but also including smaller components of C++. Our staff are not only systems administrators but also former researchers who understand how academics work and can therefore design and teach relevant courses.
Can you tell me about the decision to enable remote access?
Lindemann: Over time, Lunarc has evolved. We began with terminal-based access, offering compilers and libraries. This expanded to scientific software modules, and we now use tools such as EasyBuild to provide optimised software. Early on, we also recognised the importance of interactive desktop use. Initially, the desktop was simply an alternative to the terminal, but we later developed a system for running interactive workloads through the queuing system. Users can reserve, for example, a 256 GB node with a GPU for a day or two, and run applications directly without writing batch scripts.
This interactive environment is also valuable for experienced HPC users. It allows experimentation and development in a flexible way. For example, MATLAB users can run tests interactively without fully loading the CPUs, but still need large amounts of memory. Unlike some centres that require near-perfect efficiency from all jobs, we deliberately allow this more flexible style of use so as not to intimidate new users.
To ensure resources are used effectively, we set reasonable defaults. For instance, when running ParaView, users automatically receive four cores and partial GPU access. If they need more, they can request a full node manually. This approach lets beginners start quickly, while still offering advanced users the ability to scale up.
How is this beneficial compared to traditional HPC systems?
Lindemann: The desktop environment provides several benefits over traditional HPC access. Previously, users logged in via terminal, transferred files, and worked in their own environments, which became difficult when datasets grew too large. Now they can work directly on Lunarc’s systems with data sets of hundreds of terabytes, in the same way they would on their local machines. We provide familiar development tools such as Visual Studio Code and Qt Creator, with compilers correctly configured for our systems. This allows users to develop, compile, and run code directly where it will execute. The desktop also includes features such as file explorers, which help those less comfortable with command-line operations.
Another important feature is hardware-accelerated graphics. Researchers can run demanding visualisation software using high-performance GPUs without needing powerful workstations of their own. This eliminates the need to move large datasets back and forth and allows advanced visualisation directly on the cluster.
Remote access has further changed the way researchers interact with HPC resources. They no longer need to transfer data between systems, and they can log in from any location using a simple client. This provides both efficiency and flexibility. Around 40 percent of our users now access the system through the desktop environment, with many running visualisation software such as ParaView, Fluent, or other tools.
The main advantages of the Lunarc desktop are reduced data movement, simpler visualisation, and a familiar, accessible interface. It enables both new and experienced users to make effective use of HPC resources without unnecessary barriers.
Jonas Lindemann is the Director of LUNARC, Centre for Scientific and Technical Computing at Lund University.