中国空间科学技术

中国空间科学技术 ›› 2025, Vol. 45 ›› Issue (2): 33-41.doi: 10.16708/j.cnki.1000-758X.2025.0020

• 空间太阳能电站专题 • 上一篇    下一篇

轻质量大算力“天算计划”底层数据平台设计与应用

肖舸*， 刘艺博   

  1. 西海数道（西安）数据科技有限公司，西安710000
  • 收稿日期:2024-03-08 修回日期:2024-05-22 录用日期:2024-06-04 发布日期:2025-03-13 出版日期:2025-04-01

Design and application of lightweight high-performance data platform for “Space Computing Project”

XIAO Ge*， LIU Yibo   

  1. Xihaishudao（Xi'an） Data Technology Co.， Ltd.，Xi'an 710000， China
  • Received:2024-03-08 Revision received:2024-05-22 Accepted:2024-06-04 Online:2025-03-13 Published:2025-04-01

PDF (PC)

59

摘要： 空间太阳能开发与利用的设想正在逐步成为现实，对于星地能量传输支撑的用能方式，损耗以及对环境的影响不可避免。若能将这些原本会损耗的电能加以利用产生价值，就能有效提升空间太阳能电站电能的利用效率。在太空中建立轻质量、大算力、高效能的数据处理中心，直接使用空间太阳能电站的一部分电能，对来自卫星、地面站等数据源的空间数据进行计算处理的“天算计划”理念应运而生。天算计划的工程实现，必须解决如何在有限硬件资源下提供尽可能高的底层数据能力支撑这一问题，目前少见有效的解决方案。设计了“软超算”平台，基于高性能实时数据库软件NoSQLt，充分发挥其体量小巧、跨平台支持能力强、性能不依赖硬件的特点，直接把数据库作为数据载体，提供高并发、高吞吐的数据支持能力，以软件能力充分发挥硬件算力，从而利用部署在太空中有限的硬件资源实现高效率的数据处理功能。基于NoSQLt的诸多应用案例已经证明了软超算平台的可行性，在普通台式机和树莓派4B硬件平台上，分别实现了530万笔/秒和115万笔/秒的数值型数据的并发吞吐能力，提出了“单位吞吐能力”这一新的数据中心能耗评估指标。通过验证测试证明了NoSQLt数据库在有限硬件资源下的高效性，为将来天算计划的落地应用提供了底层能力保障。

关键词: 空间太阳能电站, 天算计划, 数据库, 时空数据, 软超算

Abstract: The space solar energy development and utilization is gradually becoming a reality. In the process of energy transfer from space to the ground， losses are inevitable. If those lost energy could be effectively utilized in space， the efficiency of energy utilization in SSPS（Space Solar Power Station）would be significantly improved. The “Space Computing Project” concept is proposed-a lightweight， large-power， and high-efficiency data processing center is established in space， and the electric energy generated by SSPS is directly used to calculate spatiotemporal data from satellites or other data sources. Providing the highest possible underlying data capability support under limited hardware resources is an engineering challenge that must be solved during the implementation of the Space Computing Project. Currently， there are few effective solutions. The “Soft Supercomputing” platform is designed. This platform is based on a high-performance real-time database software called NoSQLt. It is hardware-independent and with a small size. By using the high-concurrency and high-throughput data support capabilities provided by NoSQLt， the hardware computing capability can be fully utilized， so as to achieve efficient data processing ability with limited hardware resources deployed in space. Many application cases based on NoSQLt have proved the feasibility of Soft Supercomputing platform. For numerical data， a concurrent throughput of 5.3 million transactions per second is implemented on a regular desktop platform， and 1.15 million transactions per second on the Raspberry Pi 4B platform. A new data center energy consumption evaluation index called “Unit Throughput Capacity” is proposed.Performance testing experiments prove that NoSQLt has high performance under limited hardware resources， providing a fundamental capability guarantee for the future implementation of Space Computing Project.

Key words: space solar power station, Space Computing Project, database, spatiotemporal data, Soft Supercomputing