Research Themes
Next Generation Intelligent Data Processing Platforms
Background
PolarDB for MySQL (PolarDB-M) is a cloud-native relational database service developed by Alibaba Cloud. PolarDB-M uses the cloud-native architecture, decoupled compute and storage resources, and integrated software and hardware. Recently, PolarDB-M proposed the distributed shared memory design. It’s used in PolarDB-M’s new three-layer decoupling architecture (consisting of computation, memory, and storage layers) and in PolarDB-M multi-master design. The distributed shared memory plays an essential role in PolarDB-M’s innovations and significantly affects the system’s performance. With this design, the computation and memory nodes require a lot of information exchange, and the memory nodes need to provide sufficient computing resources to achieve cache coherence and fast recovery. Modern smart hardware, such as SmartNICs and Alibaba x-Dragon MOC cards, usually have a certain amount of computation resource. By offloading some of the memory node's computing functions to this type of hardware, it can save the memory node’s CPU resource and overall power consumption. Especially in the PolarDB-M multi-master architecture, if the lock coordination and transaction coordination functions are offloaded to this type of hardware, it can also significantly improve performance.
Target
A prototype system that leverages smart hardware to improve the performance of distributed shared memory.
Integrate such distributed shared memory (co-design with smart hardware) with PolarDB-M.
A thorough understanding of the performance bottleneck when deploying distributed shared memory in database.
Related Research Topics
Leveraging smart hardware to improve performance of distributed shared memory
To improve the performance of distributed shared memory
To save the CPU cycles in the distributed shared memory
Exploiting distributed shared memory in database
Utilize smart hardware to improve the performance of databases