Towards Agentic OS: An LLM Agent Framework for Linux Schedulers

TL;DR

OS scheduler optimization suffers from a semantic gap between application needs and kernel policies. The authors propose SchedCP, a decoupled control plane that separates AI reasoning ('what to optimize') from kernel execution ('how to observe and act'), enabling autonomous LLM agents to safely generate and deploy eBPF scheduling policies via goal-inference and policy-synthesis. The architecture includes a Model Context Protocol server with Workload Analysis Engine, Scheduler Policy Repository, and Execution Verifier, plus sched-agent, a four-agent system implementing in-context reinforcement learning to bridge workload understanding and policy synthesis. Empirical results show substantial gains across workloads, including up to ~1.79× kernel compilation speed, ~2.11× P99 latency reduction, ~1.60× throughput increase, and a 13× cost reduction relative to naive approaches, underscoring practical viability and safety of self-optimizing, application-aware OS scheduling. The work is open-sourced at https://github.com/eunomia-bpf/schedcp.

Abstract

Operating system schedulers suffer from a fundamental semantic gap, where kernel policies fail to understand application-specific needs, leading to suboptimal performance. We introduce SchedCP, the first framework that enables fully autonomous Large Language Model (LLM) agents to safely and efficiently optimize Linux schedulers without human involvement. Our core insight is that the challenge is not merely to apply a better LLM, but to architect a decoupled control plane that separates the AI's role of semantic reasoning ("what to optimize") from the system's role of execution ("how to observe and act"), thereby separating the optimization problem into two stages: goal-inference and policy-synthesis. Implemented as Model Context Protocol(MCP) server, SchedCP provides a stable interface with three key services: a Workload Analysis Engine, an evolving Scheduler Policy Repository, and an Execution Verifier that validates all AI-generated code and configure before deployment with static and dynamic analysis. We demonstrate this architecture's power with sched-agent, a multi-agent system that autonomously analyzes workloads, synthesizes custom eBPF scheduling policies, and deploys them via the sched\_ext infrastructure. Our evaluation shows that SchedCP achieves up to an 1.79x performance improvement, and a 13x cost reduction compared to naive agentic approaches, all while maintaining high success rate. By bridging the semantic gap, SchedCP democratizes expert-level system optimization and represents a step towards creating truly self-optimizing, application-aware operating systems. The code is open-sourced in https://github.com/eunomia-bpf/schedcp

Figures (2)

• Figure 1: Overall architecture of SchedCP and sched-agent. SchedCP (bottom) provides the system interface with three core services: Workload Analysis Engine, Scheduler Policy Repository, and Execution Verifier. sched-agent (top) implements the AI logic through four specialized agents (Observation, Planning, Execution, Learning) working in a closed loop. Red lines show initialization when detecting new workloads, black arrows show optimization loops, and green arrows indicate tool usage by agents.
• Figure 2: Performance evaluation of SchedCP across different workloads