In this talk, I’ll present two crash consistency works, i.e., TCCP and iDO, for systems backed with nonvolatile memory (NVM). TCCP is a hardware approach and works for nonvolatile processors of energy-harvesting systems where NVM serves as the main memory with no cache hierarchy. On the other hand, iDO is a compiler-directed approach and targets general purpose processors with deep cache hierarchy and hybrid of DRAM and NVM. I will demonstrate how these hardware and software approaches guarantee crash consistency without two much performance overhead to survive on errors such as power failures.
Changhee Jung is an Associate Professor in Computer Science at Virginia Tech. His research interests include compilers, computer architectures, and software engineering. His work has appeared in top conferences such as MICRO, PLDI, ICSE, ASPLOS, HPDC, and SC (2016 Best Student Paper Finalist and Highest Ranked Paper: 1/446), and USENIX Security. He received NSF Career Award, Google Faculty Research Award, and the Silver Prize in the SAMSUNG HumanTech Thesis Competition. Changhee received his PhD degree in Computer Science from Georgia Tech in 2013. During the three summers between 2010 and 2012, he worked with the compiler optimization team at Google.