High Performance NVM-based Storage System

Dan Feng
Professor and Dean
School of Computer Science and Technology, HUST
Location: 
SERC 306
Date: 
Thursday, October 12, 2017 - 14:00
High performance computing requires storage system with huge capacity, high bandwidth, low access latency and good energy efficiency. Emerging technologies, such as nonvolatile memories (NVMs) are playing a more and more important role in high performance computing systems for lower power consumption and better scalability. The use of NVMs, however, has brought great opportunities as well as challenges to redesign the traditional storage architecture. In this talk, the weak points of existing NVMs including PCM, STT-RAM and RRAM will be reviewed. Then, the hardware and software challenges for applying emerging NVMs in computer architecture and storage system are presented and discussed. Keeping in mind these challenges, this talk will summarize recent developments of our group in high performance NVM-based storage system from hardware level (NVM-based storage devices, read and write optimizations in NVM) to software level (Object-based NVM Management, NVM-based File System and Metadata Accelerator with NVM). Experiment results show that the advantage of NVM is enhanced and the disadvantage is avoided by the hardware and software we have designed.