Expand description
Local Scheduler — Thread-per-Core Production Queue
Implements multi-core parallelism for calculus operations by pinning isolated SLG VM loops to each logical processor. Each worker operates within a 512MB memory boundary and pulls jobs from a central lock-free ring buffer.
Architecture:
- Supervisor thread: Manages job queue and coordinates workers
- Worker threads: Pinned to specific cores, process NQuin jobs
- SPSC channels: Lock-free communication between supervisor and workers
- NVMe WAL: Persistent backing store for job state (future)
Environmental yielding:
- Thermal governor integration for power-aware execution
- Pause/resume capability via Quin state persistence
- Solar/battery power constraint handling
Structs§
- Hardware
Velocity - Hardware velocity metrics for a specific compute target
- Job
- Production queue job wrapper
- JobEstimate
- Pre-execution job estimate
- JobEstimate
Params - Job estimation parameters
- JobEstimator
- Job estimator for pre-execution time prediction
- Production
Queue - Production queue supervisor - coordinates worker threads
- Queue
Stats - Queue statistics
- Queue
Telemetry - Progress and telemetry data for the production queue
- Worker
Cell - Worker cell - isolated execution context pinned to a specific core
Enums§
- Compute
Target - Compute target for calculus operations
- JobStatus
- Job status tracking