Xkernel: Principled Performance Tunability of Operating System Kernels
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
Xkernel converts frozen Linux perf-consts into safe, in-situ knobs by synthesizing runtime handlers via Scoped Indirect Execution; it finds constants with one-shot recompile diffs plus symbolic execution, inserts handlers with Kprobe + eBPF, and enforces safety with global/thread/CS-level version atomicity. Case studies show a hotspot IO batch change yields ~50× throughput for HDD sequential reads while another setting causes 29% latency for random NVMe, and per-device or per-flow policies apply without reboot and with low overhead.
AI-written summary · about
chapters (10)
- 0:00Xkernel framework intro
- 1:03Hard-coded performance constants issue
- 3:32Runtime tunability necessity
- 5:02Limitations of existing tuning
- 5:45Xkernel’s principled tunability
- 8:21Scoped indirect execution design
- 12:19eBPF and Kprobes implementation
- 13:01Case studies and gains
- 14:50CPU and memory subsystem apps
- 15:11Conclusion and future work
