Next-generation database systems on real persistent memory
After about a decade’s wait, persistent memory (such as Intel Optane DC Persistent Memory) has recently become commercially available in 2019. These devices fundamentally change the way database systems are built: they are both byte-addressable like normal DRAM memory and persistent like flash memory and disks. They also exhibit memory-like properties and performance (albeit slower), cheaper price than DRAM yet much large capacity than DRAM. These features can bring many benefits that were not possible to achieve, namely being high-performance, low-cost and able to recover instantly. Prior research, however, has largely been simulation-driven and thus had limited applicability in today’s real PM hardware, which is shown by our preliminary study. Coupled with inherent challenges of PM, such as asymmetric read/write performance, different programming models and limited lifetime, devising a suitable database architecture becomes more challenging on real PM devices available today. This project sets out to solve these problems and aims at creating new designs for database systems on persistent memory.
View Full Project DescriptionTianzheng Wang
The Chinese University of Hong Kong
Computer science
Education
Simon Fraser University
Globalink Research Award
