This talk focuses on studies that address two challenges. The first work tackles the 'capacity' problem by enabling all classes of General Purpose-Graphics Processing Unit applications to transparently address terabytes of data residing on a non-volatile memory. The second study elaborates on the 'scalability' aspect and explores how applications should be designed for a new type of computing architecture, Emu, where threads migrate to the location where data resides.
Biography: Mehmet Belviranli is currently a post-doctoral research associate in the Future Technologies Group of the Computer Science and Mathematics Division. His research spans various areas of high-performance computing, including heterogeneous architectures, deep-memory hierarchies and performance modeling. He received his PhD in Computer Science from the University of California, Riverside in 2016 and his BS and MS degrees from Bilkent University in Ankara, Turkey.
Host: Jeff Vetter – firstname.lastname@example.org