Writing a Linux scheduler in Java with eBPF

Keynote at eBPF Summit 2024.

Transcript

2. What is scheduling?
3. Process A Process B CPU 1 ... CPU 2 ...
4. Process A Process B CPU Time
5. Process A Process B CPU A Time
6. Process A Process B CPU A Time
7. Process A Process B CPU A Time B
8. Process A Process B CPU A Time B A
9. Process A Process B CPU A Time B A A
10. Process A Process B CPU A Time B A B A
11. Process A Process B CPU A Time B A B A A
13. Runs on this machine
14. What is the performance? Good* * For a typical Java benchmark
15. Why?
16. 1.Ease of experimentation and exploration 2.Customization 3.Rapid scheduler deployments https://lwn.net/Articles/978911/ “
17. How does it work?
18. CPU sched-ext User land Kernel land Hardware Java Code C Code Byte Code Loaded attach
19. Show me the code
20. @BPF(license = "GPL") public abstract class SampleScheduler extends BPFProgram implements Scheduler, Runnable { static final long SHARED_DSQ_ID = 0; @Type static class Stats { long global; long local; } @Override public int init() { return scx_bpf_create_dsq(SHARED_DSQ_ID, -1); } }
21. @BPF(license = "GPL") public abstract class SampleScheduler { @BPFMapDefinition(maxEntries = 100) BPFHashMap<Integer, Stats> statsPerCPU; @BPFMapDefinition(maxEntries = 100000) BPFLRUHashMap<@Unsigned Integer, @Unsigned Long> enqueuesPerProcess; }
22. @BPF(license = "GPL") public abstract class SampleScheduler { @Override public int selectCPU(Ptr<task_struct> p, int prev_cpu, long wake_flags) { boolean is_idle = false; int cpu = scx_bpf_select_cpu_dfl(p, prev_cpu, wake_flags, Ptr.of(is_idle)); if (is_idle) { !/* !!... / } return cpu; } } *
23. @BPF(license = "GPL") public abstract class SampleScheduler { @Override public void enqueue(Ptr<task_struct> p, long enq_flags) { incrementStats(false); recordEnqueue(p); scx_bpf_dispatch(p, SHARED_DSQ_ID, SCX_SLICE_DFL.value(), enq_flags); } }
24. @BPF(license = "GPL") public abstract class SampleScheduler { @Override public void dispatch(int cpu, Ptr<task_struct> prev) { scx_bpf_consume(SHARED_DSQ_ID); } }
25. Why in Java?
26. Why not?
27. Testing
28. Race conditions Process A Process B Data(uid=0, gid=0, counter=10) set get ⚡
29. Deadlock Lock 1 Process A Process B own want want Lock 2 own ⚡
30. + = Better testing of concurrency
31. Can we use it?
32. Sched-ext is not yet in main-line
33. When will sched-ext be merged? I will knock wood, because I was sure it was going to be in 6.11, but yes it looks like we're now looking at 6.12. https://www.youtube.com/watch?v=nJMEuBwMD18 https://www.facesofopensource.com/linus-torvalds/ “
34. is a prototype
35. Fi#. Thanks to Dylan Reimerink sched-ext Slack Johannes Bechberger mostlynerdless.de OpenJDK Developer, SAP