Optimizing the Linux Scheduler for Gaming and Large Servers

Gaming on Linux has undergone a massive transformation, with platforms like SteamOS now running high-end titles seamlessly. However, a premium gaming experience requires more than high average frame rates; it demands the elimination of "stutter"—the micro-latencies and sudden FPS drops that break immersion. On handheld, battery-constrained devices, the challenge is even steeper: balancing peak responsiveness with aggressive power efficiency. Meeting these demands requires a departure from traditional throughput-oriented scheduling toward a more nuanced, latency-aware approach.

This talk explores how the Linux scheduler directly impacts these critical metrics. We dive into LAVD, a latency-aware scheduler engineered to minimize jitters and optimize the power-performance tradeoff. We will examine the core scheduling techniques that allow LAVD to prioritize interactivity without sacrificing throughput. Finally, we discuss the cross-domain application of these policies: how LAVD’s latency-focused design scales to large server workloads, the unique challenges of high-core-count environments, and the solutions for maintaining tail latency at scale.