- Linux patches from Tencent aim to improve handling of multiple swap devices, especially for swap tiering and layered swap.
- The new design removes the global plist and lock, introduces a percpu reader priority queue, and addresses several long-standing design issues.
- Key improvements include fair distribution of I/O to swap devices, better scalability, and support for future features like swap tiering.
- The patch series eliminates the global percpu cluster cache, which previously conflicted with concepts like swap tiering and caused performance issues.
- It also resolves problems with device rotation, contention on global locks, and unclear rotation rules, leading to a more flexible and robust system.