A new patch series extending the recently merged cache-aware scheduling subsystem in Linux is reportedly demonstrating performance gains of up to 360% for MySQL database workloads, according to a Phoronix report. The extensions aim to deepen the Linux scheduler's understanding of hardware topology, focusing optimizations on data locality.
Cache-aware scheduling marks a fundamental shift in scheduler philosophy. Moving away from a pure CPU load-balancing model, it prioritizes keeping data-intensive processes running on cores that share cache hierarchies. This minimizes costly accesses to main memory, a critical factor for performance in memory-bound applications.
The proposed extensions, building on this core concept, are reportedly showing dramatic results in targeted benchmarks. The highlighted 360% improvement for MySQL represents a best-case scenario, likely on multi-socket systems running specific query patterns that heavily leverage shared L3 cache. Such a boost would underscore the extreme sensitivity of database engines, like MySQL's InnoDB, to intelligent cache placement of their buffer pools.
The swift appearance of these advanced patches following the initial merge signals strong community confidence in this architectural direction. It positions cache-aware scheduling not as a one-time fix, but as an evolving performance paradigm essential for exploiting modern, complex CPUs with chiplet and heterogeneous core designs.
The extensions are reportedly under review by kernel developers. While their integration into a future stable release may be anticipated, definitive timelines depend on the ongoing patch evaluation. If adopted, their potential promises significant performance benefits for data center and high-performance computing environments running intensive database services.
Editor's note: This article is based on a Phoronix report. The source was inaccessible for independent verification at time of publication.
據 Phoronix 報導稱，一個延續 Linux 中近期合併的快取感知排程子系統的新補丁系列，據報正展示出針對 MySQL 資料庫工作負載高達 360% 的效能提升。這些擴展旨在加深 Linux 排程器對硬體拓撲的理解，將優化重點集中於資料的局部性。
快取感知排程標誌著排程器哲學的根本性轉變。它摒棄了純粹的 CPU 負載平衡模型，優先將處理大量資料的程序保留在共享快取層次的核心上運行。這能最大程度減少對主記憶體的高成本存取，對於記憶體受限的應用程式而言，這是效能的關鍵因素。
據報導，建基於此核心概念的擬議擴展，在針對性的基準測試中展現了顯著的成果。其中被強調的 MySQL 360% 效能提升代表了最佳情境，很可能是在運行特定查詢模式的多插槽系統上，這些模式能充分利用共享的 L3 快取。如此提升將凸顯像 MySQL 的 InnoDB 這類資料庫引擎，對於其緩衝池智慧型快取放置的極度敏感性。
這些進階補丁在初始合併後不久即迅速出現，表明了社群對此架構方向的高度信心。這將快取感知排程定位為一項不斷演進的效能範式，而非一次性修補，對於利用具備多芯粒與異構核心設計的現代複雜 CPU 至關重要。
據報導，這些擴展目前正由核心開發人員進行審核。雖然它們預期可能被整合到未來的穩定版本中，但確切時間表仍取決於持續進行的補丁評估。如果被採用，它們潛在地為執行密集型資料庫服務的資料中心及高效能運算環境帶來顯著的效能紅利。
編者註：本文基於 Phoronix 的報導。發佈時原始來源無法獨立查證。