김영석 연구실

Proposes faster but accurate performance evaluation methods as alternatives to slow cycle-level simulations. Includes GCoM for GPU performance evaluation and GCStack+GCScaler framework using stall cycle accounting.

Accelerates query processing in DBMSs through hardware acceleration techniques. Includes PID-Join for in-memory joins, SPID-Join for skew-resistant joins, FaScalSQL for GPU-accelerated queries, and DMO-DB for data movement optimization.

Designs next-generation CPU and GPU architectures for emerging workloads. Pursues software and hardware approaches through prototyping and cycle-level simulation using gem5 and GPGPU-Sim.

Implements hardware proposals on FPGAs using hardware description languages to demonstrate real-world effectiveness. Notable prototypes include DCS, DCS-ctrl, and MPC-Wrapper.