Expand description
HYDRA scheduler core: multi-source download scheduler state machine with no I/O dependencies.
Key principles:
- Dynamic range partitioning: HTTP byte ranges are tracked client-side, allowing slow or stalled connections to be repartitioned and assigned to faster connections dynamically.
- Liveness: every reachable state has an enabled transition that decreases remaining work within a bounded window.
- Safety: byte coverage invariants are strictly verified without gaps or duplicate allocations.
Safety (coverage_holds) and liveness (liveness_holds) properties are
both verified through property-based testing.
Re-exports§
pub use admission::Admission;pub use admission::Admit;pub use admission::DeltaEstimator;pub use detect::CollapseDetector;pub use detect::Health;pub use format::catalogue;pub use format::describe;pub use format::detect_format;pub use format::from_extension as from_extension_pub;pub use format::known_extensions;pub use format::Category;pub use format::Detection;pub use format::Evidence;pub use format::Format;pub use intervals::IntervalSet;pub use intervals::Range;pub use sched::greedy_concurrency;pub use sched::Action;pub use sched::Capability;pub use sched::Scheduler;pub use sched::Source;pub use sched::Stats;pub use sched::STEAL_QUANTUM;
Modules§
- admission
- Online concurrency admission: dynamically probe and scale connection counts based on measured marginal goodput.
- detect
- Fast collapse detection for connection goodput.
- format
- File-format classification from magic bytes, extension, and media type.
- intervals
- Byte-range set with a maintained coverage invariant.
- sched
- The scheduler kernel: pure state machine, no I/O, no clock, no allocation in the steady state.