Discovering US: Pete Beckman and Rajesh Sankaran

In this interview, the first in the ‘Discovering US’ series, we caught up with DISCOVER-US members Pete Beckman (Northwestern University) and Rajesh Sankaran (Argonne National Laboratory) to learn about their work and why they are part of the DISCOVER-US community.

For further information about Sage, check out the ‘Sage: A software-defined sensor network’ talk presented by Pete Beckman for DISCOVER-US.

What led you to specialize in your research area?

Pete: I’ve always been interested in fast and big computers. In the first part of my career, new supercomputer architectures and software layers were being invented each year – from new parallel languages to new massively parallel systems. Each year the systems grew faster, and I wanted to be part of that new frontier. However, as I watched instruments and sensors become ubiquitous, providing massive data streams that needed analysis, my interest turned toward taking what I had learned in high-performance computing (HPC) and shrinking it down – joining it with the instruments to build ‘edge computing’. In many ways, my career now is focused on HPC embedded with instruments.

Raj: My interest in digital electronics started at a young age. I inherited a few project books for kids from my cousin and eagerly experimented with various projects, often using whatever components I could find. I was driven by a strong curiosity to build and bring ideas to life. During graduate school, I developed a modular electronics toolkit for creating physical user interfaces using small microprocessors. Later, when faced with the challenge of integrating advanced sensors into real-time workflows, my background in embedded systems and systems science became invaluable. This marked the beginning of my journey into embedding computing along with sensors and instruments in the field and linking them to large central computing infrastructures.

What projects are you currently working on?

Pete: Our team is deeply involved in various projects aimed at advancing edge computing and AI/ML at the edge. The challenges include resource sharing at the edge, multi-tenancy, edge-cloud workflows, and programming models for programming the edge-to-HP continuum. We also leverage existing artificial intelligence / machine learning (AI/ML) methods, including generative AI, to tackle edge inference challenges in practical applications, ranging from understanding urban environments to extreme event studies and security.

Initiated in 2019, the Sage project aims to develop a national-scale cyberinfrastructure tailored for AI at the edge. It is led by the Northwestern-Argonne Institute for Science and Engineering (NAISE) and funded by the National Science Foundation (NSF). The project unites experts from several institutions, including Northwestern University, the University of Chicago, George Mason University, UC San Diego, the University of Illinois in Chicago, the University of Utah, and partners at Argonne.

Raj: Sage empowers researchers by providing adaptable, internet-connected intelligent nodes equipped with various sensors—such as cameras, microphones, and environmental monitors—that can be tailored to specific research needs through software. It also integrates with centralized cloud and HPC resources for advanced, global observations, inferences, automated workflows, and AI model development. The Sage nodes employ edge computing, enabling real-time data analysis right where it is gathered. This approach addresses the limitations of traditional sensor networks, which often struggle to manage vast data streams or rely on slow data transfers to the cloud for processing. With Sage, scientists can promptly analyse data on-site, opening up new possibilities for real-time monitoring and measurement of various urban and environmental factors.

What technical challenges have you encountered in building this infrastructure?

Raj: First, the advanced AI systems we are developing rely on electronic components that were not specifically designed for harsh, remote, outdoor environments. For instance, the central NVIDIA graphics processing unit (GPU) is intended for stable office environments with consistent electrical power, rather than the intense heat of Texas. Consequently, every design phase required meticulous considerations and rigorous testing from the outset.

Second, once deployed, these systems often cannot be physically accessed for maintenance or repairs. Similar to satellites in orbit, all diagnostics, updates, and repairs must be conducted remotely. Thus, ensuring system redundancy is crucial to prevent a single failure from compromising the entire system.

Additionally, the extreme cybersecurity risks associated with sensors and actuators, such as moving cameras, necessitate the implementation of extensive safeguards. These measures address a range of risks not typically encountered in conventional computing infrastructure.

Pete: Another challenge is multi-tenancy, where edge-computing resources are shared among various applications competing for these resources. It is essential to ensure that all applications maintain a minimum acceptable quality of output despite the rapidly changing environment. Weather conditions and the phenomena under study affect the availability of resources at the edge and the computational and memory needs of AI/ML applications.

Furthermore, we required a method to schedule jobs on the nodes remotely, allowing them to continue executing tasks despite communication loss. This involves making decisions about job scheduling based on available resources and context, all while striving to meet data and inference goals for the scientific objectives of the scheduled jobs. Each node may pursue multiple scientific objectives simultaneously.

Finally, the challenge of managing data security—at the edge, in transit to cloud and HPC resources, and within trusted computing enclaves operating on secure data—has become increasingly critical.

Why is international research collaboration important?

Pete: International collaboration plays a crucial role in driving research and innovation forward. Programmes such as DISCOVER-US allow EU researchers to spend several weeks in the US, working closely with host research groups. This period is ideal for pinpointing shared research challenges, making initial progress, and laying the groundwork for future work. Such international experiences provide valuable exposure to diverse research cultures, methodologies, and perspectives, which enhances creativity and adaptability. They also support personal development and offer a broader view of global issues.

The benefits of these collaborations are several, including access to a broader range of expertise and perspectives, shared resources, and the mobility of students and faculty. This fosters a vibrant, multicultural research environment and encourages the cross-pollination of ideas, leading to new solutions and approaches. Additionally, these collaborations facilitate the transfer of knowledge between teams. When researchers return to their home institutions, they can continue their collaborations, leveraging the progress made during the visit to achieve breakthroughs that might be challenging for either group in isolation.

Why did you join DISCOVER-US? How can it support EU and US researchers?

Raj: Science thrives on collaboration, and the diverse expertise across European universities makes partnership both straightforward and highly rewarding. Joining the DISCOVER-US community was a natural extension of our established connections with scientists in European HPC centres. This initiative facilitates month-long research visits, providing early-career, tenure-track researchers with invaluable opportunities for transatlantic collaboration. With AI now increasingly accessible on edge devices, these researchers and their students can explore innovative distributed systems, driving progress in both the EU and the US.

Pete: Our efforts with Sage and related projects are highly interdisciplinary, involving collaborations with researchers across various fields, including earth science, urban studies, soil and crop science, environmental science, data science, and visualization. This diverse research landscape offers ample opportunities to address complex problems that are not only theoretically intriguing but also hold the potential to solve significant challenges within application domains. There are numerous prospects for joint publications and collaborative grant proposals. Additionally, the NSF’s commitment to funding exchanges between the US and the EU provides a vital pathway to nurture and sustain these collaborative efforts beyond the initial European researcher visits.

Check out #HiPEACinfo73 for this interview and much more: https://www.hipeac.net/magazine/7169.pdf