Graduate student explores methods to improve performance of HPC systems Meet Mohammad Alaul Haque Monil

For Mohammad Alaul Haque Monil, the decision to intern at Oak Ridge National Laboratory (ORNL) was a no-brainer.

mohammad-alaul-haque-monil.jpg

Mohammad Alaul Haque Monil

As a doctoral student studying high-performance computing (HPC) at the University of Oregon, Monil was well aware of ORNL’s reputation as a leader in the field. ORNL is home to the nation’s most powerful supercomputer, Summit, and other state-of-the-art computing facilities.

Monil gained valuable experience in HPC research as a participant in the Advanced Short-Term Research Opportunity (ASTRO) internship program at ORNL. The ASTRO program assigns graduate students and recent master’s degree recipients to real-world projects in the basic and applied sciences, energy and environment.

Under the mentorship of Distinguished R&D Staff Member Jeffrey Vetter, Ph.D., and Senior Computer Scientist Seyong Lee, Ph.D., of the Future Technologies Group at ORNL, Monil investigated a type of computer architecture where different kinds of processors – both special-purpose and general-purpose – are put in a single computer chip to make a powerful system.

"A student can not only collaborate with world-class scientists, but also make connections with fellow interns who will be future colleagues."

This heterogeneous system, Monil explained, was designed for use by autonomous cars and drones where image processing and machine learning are essential, along with general computing needs. The processors in such systems share resources, such as system memory. When the processors are used concurrently, performance of the system may be negatively affected because they are trying to access the same memory. In computer science, this conflict is known as memory contention.

During his five-month internship, Monil investigated the memory contention in such systems and designed algorithms to achieve greater efficiency for energy and performance.

Following his appointment, Monil returned to Oregon to continue his doctoral work. Inspired by the research he conducted at ORNL, Monil developed a new direction for his dissertation, leading to further collaboration with the Future Technologies Group. “My ORNL internship not only gave me an outstanding experience but also shaped my Ph.D. research direction in a major way,” he said.

As a result of his ORNL research, Monil co-authored a paper titled MEPHESTO: Modeling Energy-Performance in Heterogeneous SoCs and Their Trade-Offs that was recently accepted to the Association for Computing Machinery’s International Conference on Parallel Architectures and Compilation Techniques. Another paper, Understanding the Impact of Memory Access Patterns in Intel Processors, has been accepted to the Memory Centric High-Performance Computing workshop of the 2020 International Conference for High Performance Computing, Networking, Storage, and Analysis.

Monil encourages other graduate students to explore opportunities at ORNL through the ASTRO program. “A student can not only collaborate with world-class scientists, but also make connections with fellow interns who will be future colleagues,” he said.

The ASTRO program is administered by the Oak Ridge Institute for Science and Education (ORISE) for the U.S. Department of Energy.