Ongoing Projects
1. Primary Data Deduplication: Data deduplication has been demonstrated to be very effective in cloud backup and archiving applications. Recently, applying data deduplication in primary storage systems, such as Flash storage and NVM, becomes popular to reduce write traffic and save cost. However, applying data deduplication in these low-latency storage devices is challenging. We proposed various solutions to investigate how to apply the deduplication technique in primary storage systems. The preliminary results have been accepted by HPCA, ICCD, SRDS, IPDPS, MSST, ACM TOS, and IEEE-TPDS/TC/TCAD.
2. Software/Hardware Co-Design for the New Storage: Recently, various new storage devices (such as Ultra-Low Latency flash, 3D NAND flash, Intel Optane PM, ZNS, et. al.) become popular in the storage systems and some of them are already deployed in the real environment. However, the traditional storage systems are designed for HDD-based storage systems, not for the new storage devices. The advantages of the new storage devices are not fully exploited by the upper applications. We are in the process to design new File systems, I/O schedulers, and data placement techniques to exploit these new device characteristics. The preliminary results have been accepted by ICS, ICCD, MSST, DATE, IPDPS, Cluster, ACM-TOS, and IEEE-TPDS/TC/TCAD.
3. High Available Storage Systems: Data centers are consisting of hundreds and thousands of electronic components, including NICs, CPUs, Disks, and so on. Device failures in such an environment become common cases rather than exceptions. Thus fast recovery from device failures, including partial and complete failures, becomes critical for the availability of large-scale storage systems. We have proposed some efficient data recovery algorithms and systems, please see our published papers in FAST, ICDCS, LISA, IPDPS, and IEEE-TPDS/TC.
We would like to thank our sponsors for their generous support over the years, including:
