Sungjoo Yoo
Seoul National University, Korea
Reinforcement learning-based optimization for solid state disks
Download SlidesAbstract
Garbage collection (GC) on solid state disk can incur long tail latency which is harmful for servers and real-time embedded systems. In order to reduce the GC-induced latency, we propose applying reinforcement learning (RL). The RL agent dynamically learns storage access behavior including idle time and, on every request service, selects a suitable action of page copy operation including no copy while exploiting idle time to hide GC latency. We present a Q-table based solution to show the feasibility of RL approach to this problem. We also present a novel idea called Q-table cache which utilizes temporal locality in SSD accesses and manages a tiny cache structure of Q-table. The Q-table cache increases the effective size of Q-table, at a very small memory cost of the cache structure, thereby offering further reduction in long tail latency.
Biography
Sungjoo Yoo received Ph.D. from Seoul National University in 2000. From 2000 to 2004, he was researcher at system level synthesis (SLS) group, TIMA laboratory, Grenoble France. From 2004 to 2008, he led, as principal engineer, system-level design team at System LSI, Samsung Electronics. From 2008 to 2015, he was associate professor at POSTECH. In 2015, he joined Seoul National University and is now full professor. His current research interests are software/hardware co-design of deep neural networks and machine learning-based optimization of computer architecture.