QiMeng-PerceptOS: Semantic-Aware Kernel Optimization for OS-Intensive Workloads via Hardware-Software Alignment

Optimizing OS kernels for specific applications is vital for peak performance, yet existing LLM-based methods struggle with a semantic mismatch between generalized reasoning and low-level system behaviors. As a result, these static, open-loop approaches suffer from runtime blindness, configuration fragmentation, and search drift, ultimately failing to unlock the system’s performance potential. To address this, we propose QiMeng-PerceptOS, an autonomous framework that shifts the paradigm to perception-driven tuning. QiMeng-PerceptOS integrates: (1) a Perception Module that aligns raw telemetry into high-fidelity semantic fingerprints; (2) a Global Search Module utilizing a Bi-level Hierarchical Induction Tree (BHIT) for global navigation and efficient pruning; and (3) a Posterior Enhancement Module to suppress hallucinations via trajectory synthesis. Experiments across diverse workloads show that it achieves significant performance breakthroughs by optimizing kernel configurations, reaching 296.6% of default Redis throughput and surpassing SOTA baselines by 32.6% within only 15 iterations. By establishing a perception-driven closed-loop, QiMeng-PerceptOS provides new insights for fully automated, large-scale system optimization.

Modern operating systems contain over 15,000 configuration options, yet tuning them for the specific software running on top remains extremely difficult. Existing AI-based methods act like blind observers; they fail to connect with the computer's actual runtime state, leading to inefficient or invalid configurations that waste significant testing time. We introduce the QiMeng-PerceptOS autonomous framework, which grants AI the semantic perception ability to optimize OS configurations. This fully automated closed-loop framework (1) translates complex hardware and software data into semantic information that the AI can understand, (2) establishes a global map of configurations to enable precise navigation of the vast search space, and (3) re-examines historical trajectories to avoid errors made by AI. By bridging this semantic gap, QiMeng-PerceptOS drastically improves the performance of software running on the OS.