Paper Title: Alibaba's In-Storage Computing Architecture and Ecosystem

Paper Abstract: With the rapid growth of the data-driven applications, including AI, big data, content delivery, database, and machine learning etc., the computation of the high volume data becomes an important customer requirement in addition to the storage. Traditional architectures conduct the computation through the CPU, and thus need to transfer data from storage to memory first, resulting in high-volume data transfers which cause significant I/O, network and CPU overhead. Adding computational capabilities to the storage devices avoids the performance-expensive transfer of data prior to computation. This talk will describe an architecture and its specifications for ThinkStor, Alibaba's In-storage computing platform, including its hardware and software architecture, potential applications, and its ecosystem.

Paper Author: Feng Zhu, Staff Engineer, Alibaba Cloud

Author Bio: Feng Zhu is a staff engineer and storage architect at Alibaba Infrastructure Service Group (AIS), focusing on architecting and developing of the storage hardware/software for hyperscale data center applications. His primary research interests are on the software and hardware co-optimization for open channel SSDs. Before joining Alibaba, Feng was working for Intel Non-volatile Solution Group (NSG) as a NAND Media-System architect and Firmware Engineering manager. His primary focus at Intel was on the path-finding and productization of SSD Flash Translation layer (FTL) and Media Management modules. Feng was the leading Media-System architect for Intel P4510 SSD. Feng obtained his PhD degree on Electrical and Computer Engineering from University of Texas at Austin and bachelor degree of Microelectronics from Tsinghua University, Beijing.