One of the key challenges of research in the field of computer science is assembling the appropriate infrastructure needed for scientists to successfully carry out their work. One of the founding principles of cloud computing, fully upheld by ELKH Cloud [1], is enabling users to provision substantial computing resources swiftly and effortlessly. ELKH Cloud aims to support members of the scientific community in the establishment of their desired research environments as one of the leading Infrastructure as a Service (IaaS) providers of Hungary.

The difficulties researchers face however, do not cease with the convenient access to computing resources, as studies in the field of computer science often require complex system architectures and an intricate mesh of software and services. Consequently, ELKH Cloud intends to support the work of researchers beyond the scope of infrastructure, and further promote the quickrealization of new ideas by providing Platform as a Service (PaaS) level components.

Such components are reference architectures [2], provided by ELKH Cloud to ease the deployment of complex software systems, a demanding task faced by many members of the scientific community. These are architectural blueprints composed of proven solutions, built with best practices in mind, to be applied in a variety of sub-fields of computer science. Reference architectures are based on the principles of the Infrastructure-as-Code (IaC) methodology, which in combination with the orchestration and configuration management tools applied, enables scientists to perform the automated deployment of their research environments in a matter of minutes. Similarly to ELKH Cloud itself, reference architectures are built based upon open-source technologies, furthering the propagation of Open Science in Hungary.

ELKH Cloud is developed by researchers who work closely together with other members of the scientific community, and hence have a good understanding of their needs and practices. The numerous different reference architectures made available on ELKH Cloud cover prominent areas such as deep learning, high performance computing, workload management, container orchestration, and several others. All this is realized by relying on some of the most prevailing tools, technologies, and platforms, thus ensuring that researchers on ELKH Cloud can smoothly carry out their work with an up-to-date, suitable toolset.

While ELKH Cloud is a general-purpose infrastructure, reference architectures enable the utilization of its resources on a highly efficient level in various specific use cases. For instance, distributed deep learning is an increasingly important topic nowadays, and it is one of the areas in which the question of infrastructure is a crucial one due to the considerable resource requirements of the process. Using the Horovod reference architecture [4], distributed training can be performed on ELKH Cloud with great scaling efficiency [3]. Beyond the existing reference architectures, ELKH Cloud also provides technical consultation services for its users, aiding them in efficiently utilizing the cloud, or even supporting their work with specifically granted resources, be it hardware or software. This can lead to the creation of new reference architectures precisely tailored to the given research project. Platform as a Service (PaaS) level components and the principles applied play a crucial role in elevating the usability of the platform, and empowering research projects to utilize its resources to their full potential.

[1] ELKH Cloud, Cloud services for national and international research projects
 Available at https://science-cloud.hu/en

[2] ELKH Cloud, Reference architectures
 Available at https://science-cloud.hu/en/reference-architectures

[3] A. Farkas, K. Póra, S. Szénási, G. Kertész, R. Lovas. “Evaluation of a distributed deep learning framework as a reference architecture for cloud environment,”
 IEEE 10th Jubilee International Conference on Computational Cybernetics and Cyber-Medical Systems (ICCC 2022)

[4] Science Cloud, Horovod Reference Architecture
 Available at https://git.sztaki.hu/science-cloud/reference-architectures/horovod