Alibaba Innovative Research (AIR) > Next Generation Intelligent Data Processing Platforms
Research on the Core Issues of the Database based on New General-Purpose Processors

Research Themes

Next Generation Intelligent Data Processing Platforms


Alibaba Cloud's PolarDB is a cloud-native relational database that is compatible with MySQL and PostgreSQL. It is designed to deliver high-performance, scalability, and availability for mission-critical applications. PolarDB supports automatic scaling, backup, and disaster recovery. It also offers features such as global synchronization, read/write splitting, and cross-region backup and recovery. Overall, PolarDB is a flexible and reliable database solution for businesses that require high levels of performance and availability in their database environment.

"Back to Basic" is one of the Alibaba Cloud's most important strategy this year. Through CIPU, we are focusing on deep foundation and strengthening the cloud computing core. The general-purpose processor, represented by Yitian 710, is the most important factor influencing the future direction of cloud computing. However, these new processors also introduce many new challenges and opportunities.

In addition to full-stack correctness adaptation (such as weak memory order under the ARM architecture), there are often some performance issues (performance of individual physical cores, severe performance decay across node/NUMA, etc.). Furthermore, in terms of database business scenarios, being able to provide some business-specific feature requirements for the subsequent development of Yitian series processors, doing deeper software-hardware codesign research, may bring database capabilities that are different from those of other databases.


Patents and papers at top system conferences.

Code including research prototype, database system kernel modules, etc.

A thorough understanding of the potential optimizations of databases on new general-purpose processors

Related Research Topics

  1. Correctness adaptation/performance optimization based on weak memory order.
  2. Solve the problem that cross-node/NUMA can become a significant bottleneck.
  3. Incorporate specific business scenarios from the database, and deepen the collaboration between software and hardware design.

Scan QR code
关注Ali TechnologyWechat Account