Redefining Performance Monitoring for Energy-Efficient Computing

Researchers at the Maine College of Engineering and Computing (MCEC) are paving the way for improved performance monitoring in next-generation computer processors, addressing critical challenges posed by hybrid designs.

Computer processors, often referred to as the “brains” of modern devices, are made up of cores—individual units capable of handling computing tasks. Traditionally, processors used identical cores to streamline design and programming. However, hybrid processors combine powerful cores designed for demanding tasks with energy-efficient cores for simpler functions, delivering faster performance and longer battery life in everyday devices, impacting everything from gaming and streaming to scientific research and business operations.

Heterogeneous processors, which combine high-performance cores (P-cores) and energy-efficient cores (E-cores), are transforming the computing landscape, offering improved energy efficiency and performance. However, these advancements present new complexities in measuring performance accurately and efficiently across different core types.

The Performance API (PAPI) is a widely used software library that enables researchers and developers to measure and analyze computer performance. It provides access to hardware performance counters, which are built into processors to track metrics like processing speed, memory usage, and power efficiency. PAPI is an essential tool in optimizing software for high-performance computing, helping industries and scientists improve their applications and systems.

Vincent Weaver, associate professor of electrical and computer engineering at UMaine and graduate student Willow Cunningham are leading efforts to adapt PAPI to support these hybrid systems. Their research focuses on extending PAPI to ensure it can effectively track and analyze performance data on heterogeneous processors, a critical need as these chips become standard in desktops, laptops, and high-performance computing systems.

“Heterogeneous processors are reshaping how we think about computing, but they require tools that can handle their complexities,” Weaver said. “By enabling PAPI to support hybrid systems, we’re ensuring researchers and developers have the resources to optimize their software for this emerging technology.”

The team’s research highlights the potential for tools like PAPI to mitigate challenges in task scheduling and performance monitoring, laying the groundwork for smoother adoption of hybrid processors in advanced applications. With their findings, MCEC researchers are positioning UMaine as a leader in addressing the technological demands of modern computing.

This research highlights MCEC’s role in advancing sustainable and efficient computing solutions for the future. Supported by the National Science Foundation through the four-year grant “Scalable Performance and Accuracy Analysis for Distributed and Extreme-Scale Systems,” this project is a collaboration with researchers from the University of Tennessee and the University of Texas El Paso. The grant focuses on enhancing performance measurement tools, particularly the widely used Performance API (PAPI), for which the MCEC team leads core development. By improving these tools, the team aims to help researchers optimize their code to run faster and more efficiently on supercomputers, maximizing the value of time on systems with millions of processors.

Their research, titled “Performance Measurement on Heterogeneous Processors with PAPI,” was presented at the Workshop on Programming and Performance Visualization Tools (ProTools 2024) in Atlanta, Georgia. The workshop is part of the ACM/IEEE Supercomputing 2024 conference, the largest and most esteemed event for high-performance computing research. 

The Maine College of Engineering and Computing is a signature initiative of UMS TRANSFORMS, a $320M investment by the Harold Alfond Foundation to drive the innovations and technical workforce critical to advancing Maine’s economy. 

Contact: Taylor Ward, taylor.ward@maine.edu