Speaker
Description
China's High Energy Photon Source (HEPS) will complete facility construction and commissioning by the end of 2025. Data acquisition and analysis have already begun. The 14 beamlines of the first phase of HEPS will generate approximately 300PB of raw data annually, with further expansion expected in the future. This not only poses significant challenges for the reliability and read-write performance of data storage but also demands more complex access patterns. To address these challenges, several measures have been implemented, such as: adopting a "hot-warm-cold" three-tier storage architecture and providing users with long-term data download web services; deploying an all-flash, fully symmetric distributed file system as the "beamline storage," which utilizes a proprietary RDMA-based protocol to handle high-bandwidth data read-write operations; utilizing open-source storage software (Lustre) to build a highly available large-scale data storage system for medium-term to long-term data preservation, ensuring that single hardware failures do not affect data services; and developing various AI-based operational tools, among others.