Skip to content

ROADMAP.md

Latest commit

 

History

History
117 lines (94 loc) · 5.35 KB

ROADMAP.md

File metadata and controls

117 lines (94 loc) · 5.35 KB

Roadmap

v0.4.0 — Production Hardening (DONE)

Core Affinity Helpers (DONE)

  • affinity::pin_to_core(), pin_to_core_id(), available_cores(), core_count()
  • no_std compatible via core_affinity2 crate (no std dependency)
  • Re-exports CoreId for direct use
  • NUMA-aware ring allocation (v0.5.0)
  • Benchmark with pinning enabled vs disabled on multi-socket Xeon (v0.5.0)

Backpressure Mode (DONE)

  • channel_bounded(capacity, watermark) — non-lossy bounded channel
  • Publisher::try_publish() → Err(PublishError::Full(value)) when ring is full
  • Per-subscriber cursor tracking with publisher-side min-scan on slow path
  • Zero overhead on default lossy channel() (backpressure is Option::None)
  • 9 backpressure-specific tests including cross-thread correctness

Wait Strategies (DONE)

  • WaitStrategy enum — fully no_std, no OS primitives
  • BusySpin — bare spin, zero wakeup latency
  • YieldSpin — core::hint::spin_loop() (PAUSE on x86, YIELD on ARM)
  • BackoffSpin — exponential backoff with increasing PAUSE bursts
  • Adaptive { spin_iters, yield_iters } — three-phase escalation (default)
  • recv_with(strategy) on both Subscriber and SubscriberGroup

v0.5.0 — HFT-Grade Infrastructure (DONE)

Memory Control (DONE)

  • Huge page allocation (2MB) via mmap + MAP_HUGETLB (mem::mmap_huge_pages)
  • Publisher::mlock() — lock ring pages in RAM (prevent swap/page faults)
  • Publisher::prefault() — pre-fault all ring pages on demand
  • Cache line alignment verified at compile time (const_assert on Slot)
  • NUMA-aware allocation via set_mempolicy (v0.6.0)

Observability (DONE)

  • Subscriber::total_received() / total_lagged() / receive_ratio()
  • SubscriberGroup::total_received() / total_lagged() / receive_ratio()
  • Publisher::sequence() — current sequence for lag computation
  • Optional RDTSC-stamped latency histogram (v0.6.0)
  • Prometheus / StatsD export (v0.6.0)

Examples (DONE)

  • examples/pinned_latency.rs — core-pinned RDTSC latency measurement
  • examples/backpressure.rs — reliable order fill pipeline
  • Example: DPDK → Photon Ring pipeline (v0.6.0)
  • Example: Solarflare ef_vi → Photon Ring fan-out (v0.6.0)

v0.7.0 — Advanced Patterns (DONE)

Multi-Producer Support (DONE)

  • channel_mpmc() — CAS-based sequence claiming for multi-producer
  • MpPublisher<T> — Clone + Send + Sync, &self publish
  • Ordered cursor advancement (consumers see messages in sequence)
  • Benchmark MPMC vs SPMC to quantify the CAS overhead (future)

Consumer Topologies (DONE)

  • Pipeline: A → B → C — examples/pipeline.rs
  • Diamond: A → {B, C} → D — examples/diamond.rs

NUMA-Aware Allocation (DONE)

  • mem::set_numa_preferred(node) — set memory policy for ring allocation
  • mem::reset_numa_policy() — reset to system default

Typed Topic Bus (DONE)

  • TypedBus with publisher::<T>(name) / subscribe::<T>(name)
  • Type-erased storage via Box<dyn Any>, panics on type mismatch

v2.5.0 — Eliminating Cons (DONE)

Arbitrary Capacity (DONE)

  • Lemire fastmod for non-power-of-two capacities
  • is_pow2 branch for zero regression on power-of-two
  • 15 new tests including exhaustive verification

Pipeline Wait Strategy (DONE)

  • then_with(), then_a_with(), then_b_with() APIs
  • Backward-compatible (existing APIs delegate with default)

Derive Macro Hardening (DONE)

  • #[photon(as_enum)] attribute for explicit enum marking
  • Compile error for unrecognized field types

Companion Crates (DONE)

  • photon-ring-async — runtime-agnostic async wrappers
  • photon-ring-metrics — framework-agnostic observability

Formal Verification (DONE)

  • Loom model for MPMC cursor protocol (4 scenarios)
  • Full loom integration with source via cfg(loom) (future)
  • proptest / cargo-fuzz property-based testing (future)

v0.8.0 — Research & Formal Methods (DONE)

Platform-Specific Optimizations (DONE)

  • ARM WFE instruction in BackoffSpin and YieldSpin (aarch64)
  • UMWAIT/TPAUSE implemented as MonitorWait/MonitorWaitFallback with runtime CPUID detection
  • SPMC vs MPMC benchmark comparison (2.8 ns vs 11.7 ns, 4.2x CAS overhead)
  • RISC-V WRS (wait-on-reservation-set) when available

Formal Verification (DONE)

  • TLA+ model of the seqlock-stamped ring protocol (verification/seqlock.tla)
  • NoTornRead safety property verified in spec
  • MC.tla model config + README with TLC instructions
  • atomic-slots feature: Formal soundness gap closed. Seqlock data race (UB under Rust abstract machine) eliminated by decomposing T: Pod payloads into [AtomicU64; N] stripes. Zero performance regression on x86-64 (identical MOV instructions). Miri-passable. See docs/research-seqlock-alternatives.md for the constraint-anchored analysis that produced this design.
  • Loom-based concurrency testing (when loom supports seqlock patterns)
  • Property-based testing with proptest

Academic Publication (DONE)

  • Technical report outline (docs/technical-report.md)
  • Structured for comparison with Disruptor, Aeron, Chronicle Queue
  • Full paper text (future)
  • Submit to a systems venue (future)