Creating Ecosystems to Increase Utilization and Programmability of Extremely Heterogeneous Architectures

Abstract

The slowdown of Moore’s law caused a spree in architectural diversity over the last decade and, as a result, processing units targeted at accelerating a specific computational domain are becoming commodities. The system heterogeneity reaches extreme ends in state-of-the-art embedded and high-performance computing platforms where dozens of different type of accelerators share the same interconnect.
In the near future, meeting the rapid increase in computational needs without breaking the power and latency limits will only be possible with heterogeneous ecosystems that are able to orchestrate the application execution across all available compute units while fully exploiting their domain-specific abilities via common programming paradigms.
In this talk, I will give highlights from my research that contribute towards increasing programmability and utilization of heterogeneous systems. I will also talk about my future research plans on building a comprehensive ecosystem of programming tools, runtime systems, hardware organization, and performance models for extremely heterogeneous architectures.

Biography

Mehmet E. Belviranli is a computer scientist in Future Technologies Group at Oak Ridge National Laboratory. His main research interests have been targeted at increasing resource utilization in heterogeneous systems. He developed runtime systems, scheduling algorithms, analytical models, extended memory spaces, programming abstractions and systems & architecture level support for more efficient heterogeneous computing. Mehmet received his Ph.D. degree in Computer Science and Engineering from University of California, Riverside in 2016 and he continued his research at Oak Ridge National Laboratory since then.