016229cbcf
In some games (e.g., Elden Ring), it was observed that preemption happens much less frequently. The reason is that tasks' runtime per schedule is similar, so it does not meet the existing criteria. To alleviate the problem, the following three tunables are revised: 1) Smaller LAVD_PREEMPT_KICK_MARGIN and LAVD_PREEMPT_TICK_MARGIN help to trigger more preemption. 2) Smaller LAVD_SLICE_MAX_NS works better especially 250 or 300Hz kernels. 3) Longer LAVD_ELIGIBLE_TIME_MAX purturbes time lines less frequently. Signed-off-by: Changwoo Min <changwoo@igalia.com> |
||
|---|---|---|
| .. | ||
| src | ||
| .gitignore | ||
| build.rs | ||
| Cargo.toml | ||
| LICENSE | ||
| meson.build | ||
| README.md | ||
| rustfmt.toml | ||
scx_lavd
This is a single user-defined scheduler used within sched_ext, which is a Linux kernel feature which enables implementing kernel thread schedulers in BPF and dynamically loading them. Read more about sched_ext.
Overview
scx_lavd is a BPF scheduler that implements an LAVD (Latency-criticality Aware Virtual Deadline) scheduling algorithm. While LAVD is new and still evolving, its core ideas are 1) measuring how much a task is latency critical and 2) leveraging the task's latency-criticality information in making various scheduling decisions (e.g., task's deadline, time slice, etc.). As the name implies, LAVD is based on the foundation of deadline scheduling. This scheduler consists of the BPF part and the rust part. The BPF part makes all the scheduling decisions; the rust part loads the BPF code and conducts other chores (e.g., printing sampled scheduling decisions).
Typical Use Case
scx_lavd is initially motivated by gaming workloads. It aims to improve interactivity and reduce stuttering while playing games on Linux. Hence, this scheduler's typical use case involves highly interactive applications, such as gaming, which requires high throughput and low tail latencies.
Production Ready?
This scheduler could be used in a production environment where the current code is optimized. The current code does not particularly consider multiple NUMA/CCX domains, so its scheduling decisions in such hardware would be suboptimal. This scheduler currently will mainly perform well on single CCX / single-socket hosts.