Daniel Araújo de Medeiros

Sustainable High-Performance Computing for a Better World

Unless you are a zombie or a night owl, you are unlikely to be using the internet at night as you are sleeping. This essentially means that the demand for internet services - such as the ones provided by major companies like Google, Amazon and Netflix - has a broad variation during a day.

Such services are usually "in the cloud": the user does not care where the servers are actually located as long as it provides a reliable service. That said, the computer systems at the datacentres in which these services are hosted are already designed with cost effectiveness in mind. As electrical consumption is a major cost for datacentres, even small savings can have a huge financial impact. However, there are constraints on how optimized the hardware can be without severely impacting performance and quality of service perception by the user.

An alternative is to look into software engineering. Cloud applications are now developed with features such as reability and elasticity in mind, and the latter means that in peak demand you would have more servers online and at the bottom demand, you could have your service scaling down. This approach gives an extreme efficiency to the system, and the scaling happens without any noticing from the user. This is only possible because most of cloud applications are loosely-coupled: the servers do not depend on each other to execute. In practice, you are able to remove one server or increase one withou any major problems.

Things looks a lot more different when we look at high-performance computing systems. System design for HPC applications often do not take into account the energy consumption: rather, full performance is usually regarded as more important than performance per watt. Furthermore, the applications use cases for these systems are mostly from scientific domains such as molecular dynamics, fluid simulations, or quantum mechanics, and most of them are tight-coupled: some calculations can only proceed when all the involved parties are done. Traditional HPC systems also do not include many equivalent cloud features such as the elastic scaling seem above and, as not applications are equal and resources are scarce, this can seriously compromise the efficiency of the system.

My research interests are aligned towards understanding how can we bring the benefits from cloud systems, which are heavily optimizated for efficiency, to HPC applications, which are optimized for performance. This includes but is not limited to the deployment of HPC applications on the cloud, containerization, workflows systems, monitoring/profiling, and the usage of different computational architectures such as ARM, GPUs and FPGAs.

International Collaborations

Multidisciplinary collaborations are frequent for me. The two EU-funded projects I currently participate are composed by partners that includes both top-tier universities and extremely innovative companies that aim to the goal of making Europe sovereign in its high-performance computing stack.

The IO-SEA project aims to provide I/O solutions that are suitable for the exascale (or 1,000,000,000,000,000,000 operations per second), the next frontier for JUPITER - Europe's next supercomputer. IO-SEA develops research products such as a semantic storage API, an ephemerous storage systems, a workflow manager with focus on IO performance, among other products that are likely to be used in the following EUPEX project. During the IO-SEA project, I have so far coordinated and authored one full deliverable report, with a second one due next year.

OpenCUBE focuses on developing a cloud-based stack for the upcoming high-performance European Processor, developed by one of its partners (SiPearl). This project intends to bring the benefits of cloud-based systems to high-performance applications that are not easily accessible in traditional HPC systems, such as malleability, workflows, and disaggregated memory. My role concerns in designing system-level components and extending existing open source tools, such as Kubernetes,  to support the hardware.

During my masters, I have formely collaborated with people from two international institutions, namely the Barcelona Supercomputing Center and the University of Pittsburgh.

Research Output

I have led several publications in traditional venues in the cloud computing and high-performance computing fields, such as CCGrid and SBAC-PAD. At KTH, my main supervisor is Dr. Ivy Peng, with Dr. Stefano Markidis and Dr. Pawel Herman having supporting leads in my research. I take an active role in supporting the organization of research-related events, such as the PDC Summer School 2023 (Teaching Assistant) and the International Conference on Parallel Processing 2024 (Web Chair).

In geosciences, my selected highlights are the co-autorship of some chapters in a book (link in portuguese) related to the mapping of phosphates minerals in Brazil and the supporting role, as a geophysicist, of the ellaboration of multiple geological maps in the state of Bahia, Brazil (e.g., 1, 2; links in portuguese).I also had a minor role in the mineral evaluation report of the Palmeirópolis Mineral Complex (link in portuguese), later auctioned by the Brazilian Federal Government for over 50 million dollars to an Australian company.

Leadership

In 2019, I have been awarded a travel grant by the Brazilian Internet Steering Committee to be one of the representatives at the United Nation's Internet Governance Forum 2019, in Geneva (Switzerland). During my time in Brazil, I often volunteered into education- or environmental-related NGOs, such as Fundação Estudar (link in portuguese) and Greenpeace (link in portuguese).

As of July 2023, I am the elected head of the PhD Chapter at KTH, and have a seat both at the KTH University Board and at the THS Student Union Council. We are extremely committed towards excellence in research education, inclusion, diversity and the doctoral students well-being.

I have also been planning, managing and coordinating high-profile events intended for doctoral students, such as the PhD Gasque 2023 (budget: 110K SEK, 101 students), the doctoral ski trip 2023 (budget: 180K SEK, 193 students), and the upcoming doctoral musical festival 2024 (budget: 130K SEK, 1000 students expected).

Education and Work Experience

I hold a Bachelor's degree in Geophysics from the Federal University of Rio Grande do Norte (link in portuguese), a Master's degree in Computer Science from the Federal University of Bahia (link in portuguese), and a post-graduate degree in Business Administration by the Fundação Getulio Vargas (link in portuguese), an institution that is consistently ranked as one of the best think tanks in the world. I am currently pursuing my doctoral studies at KTH Royal Institute of Technology and intend to acquire a licenciate degree from the same university halfway.

From 2014 to 2021, I worked at the Geological Survey of Brazil (link in portuguese), a state-owned enterprise, as a Researcher/Geophysicist. I speak fluently Portuguese and English, and have acquired an intermediate knowledge in Swedish and Spanish.

Disclaimer

The views in this page are my own and do not represent KTH nor the projects I am affiliated. For direct enquiries regarding any of my management roles related to KTH or the PhD Chapter, please send a message directly to .

Images credits: IX.br, Morné Kruger, and companies for their own logo.