DBMS for NVRAM
With the byte-addressability and durability of emerging non-volatile memory, such as spin-transfer torque MRAM (STT-MRAM), ACID principles can be enforced in a more efficient way in DBMS. We develop Eternity , a novel database management system for byte-addressable NVRAM.
Android I/O Stack
SQLite is a server-less database engine embedded in Android, which is often blamed for its poor interaction with EXT4 file system. In this project, we improve the I/O performance of SQLite.
High Performance Machine Learning
We explore the opportunity to leverage high performance computing technologies to accelerate distributed machine learning. This project is in its infancy, but will hopefully be one of the main research themes in our lab.
IEEE/ACM CCGRID 2018, "CAVA: Exploring Memory Locality for Big Data Analytics in Virtualized Clusters"
USENIX FAST 2018, "Endurable Transient Inconsistency in Byte-Addressable Persistent B+-Tree"
ACM TOS SI, "clfB-tree: Cacheline Friendly Persistent B-tree for NVRAM"
JPDC, "Co-Processing Heterogeneous Parallel Index for Multi-Dimensional Datasets"