release: bump version to 0.1.1

- Update CHANGELOG.md with v0.1.1 release notes and performance metrics
- Update README.md with new benchmark numbers and optimization highlights
- Update USAGE.md changelog section with v0.1.1 details
- Bump Cargo.toml version 0.1.0 → 0.1.1

Author: bravo1goingdark

CHANGELOG.md

@@ -5,33 +5,35 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [Unreleased]
+## [0.1.1] - 2026-03-31
 
-### Added
+### Performance
 
-- `# Example` code blocks on every public method and type (50 doc tests)
-- `CHANGELOG.md`
+- Up to **27% faster** concurrent inserts (4 threads) and **20% faster** single-threaded sequential inserts
 - O(1) `memory_usage()` via atomic running total (was O(N shards))
 - Adaptive backoff (`core::hint::spin_loop()`) on CAS failure to reduce cache-line bouncing
-- Iterator prefetching of next-next node for better cache behavior during scans
-
-### Changed
-
-- `insert_inner()` returns `(InsertResult, usize)` — exact allocation size tracked per insert
 - `insert_batch()` optimized: single sealed check, single arena lookup, no per-insert overhead
 - `init_node()` uses bulk u64 header writes (3 stores) instead of 8 individual stores
 - `compare_keys()` uses `unwrap_unchecked()` behind length guards to eliminate bounds checks
-- `#[inline(always)]` on all hot-path functions (`compare_keys`, `prefetch_read`, `init_node`, `tower_load`, `tower_store`, `tower_cas`)
+- Conditional `record_alloc` — atomic tracking only when memory limits are configured
+- Removed unnecessary prefetch from `get`, `delete`, `seek` paths (extra tower_load hurt cache-resident workloads)
+- Kept lookahead prefetch in `find_less` (beneficial for insert-heavy workloads with larger datasets)
+- `#[inline(always)]` on all hot-path functions
 - Release profile: `lto = "fat"`, `strip = true` for maximum optimization
+
+### Changed
+
+- `insert_inner()` returns `(InsertResult, usize)` — exact allocation size tracked per insert
 - `ConcurrentArena`: shard cache `Vec` → `HashMap` → 8-entry inline array with `Drop` cleanup
-- `Cargo.toml`: add `[profile.release]` with `codegen-units = 1` + `lto = "thin"` (10-30% perf)
 - `TowerPtr::is_null`: skip pointer mask, check raw value directly
 - `#[inline]` on all `Iterator::next` impls (`Iter`, `SnapshotIter`, `Cursor`)
 - `#[cold]` on `should_seal` (cold path hint for branch prediction)
 - `compare_keys`: use `from_be_bytes` instead of `from_ne_bytes` + `swap_bytes`
-- `Cargo.toml`: updated `repository` URL to `themankindproject/fastskip`
-- `Cargo.toml`: added `documentation`, `homepage`, `readme`, `exclude` fields
-- `README.md`: fixed build badge URL to `themankindproject/fastskip`
+
+### Added
+
+- `# Example` code blocks on every public method and type (50 doc tests)
+- `CHANGELOG.md`
 
 ### Removed
 

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "fastskip"
-version = "0.1.0"
+version = "0.1.1"
 edition = "2021"
 authors = ["Ashutosh Kumar <kumarashutosh34169@gmail.com>"]
 description = "Lock-free arena-backed skip list memtable for LSM-tree storage engines"

Cargo.toml

@@ -210,29 +210,34 @@ Benchmarks on AMD Ryzen 5 3600 (6-core, 3.6GHz). All numbers from `cargo bench`.
 
 | Threads | Throughput | Latency/insert |
 |---------|------------|----------------|
-| 4 | **8.4M ops/s** | 119ns |
+| 4 | **9.4M ops/s** | 106ns |
 | 8 | **7.8M ops/s** | 128ns |
 
 ### Single-threaded latency
 
 | Operation | fastskip | BTreeMap | HashMap |
 |-----------|----------|----------|---------|
-| insert (random) | 221ns | 90ns | 21ns |
-| get hit | 54ns | 33ns | 17ns |
-| get miss | 24ns | 78ns | 16ns |
-| cursor seek | 51ns | 551ns | — |
+| insert (seq, 10K) | 136µs | 1.37ms | 171µs |
+| insert (rand, 10K) | 2.2ms | 1.5ms | 212µs |
+| get hit | 80ns | 43ns | 20ns |
+| get miss | 38ns | 91ns | 16ns |
+| cursor seek (1K) | 36ns | 397ns | — |
+| cursor seek (10K) | 60ns | 5µs | — |
 
 fastskip trades single-threaded speed for lock-free concurrent writes — BTreeMap and HashMap require external locking for multi-threaded access, which destroys throughput.
 
-### Recent Optimizations
-
-- **O(1) `memory_usage()`** — atomic running total instead of iterating all shards
-- **Bulk header writes** — `init_node()` uses 3×u64 stores instead of 8 individual stores
-- **Iterator prefetching** — lookahead prefetch of next-next node during scans
-- **Adaptive backoff** — `spin_loop()` hint on CAS failure reduces cache-line bouncing
-- **Bounds-check elimination** — `unwrap_unchecked()` in `compare_keys()` behind length guards
-- **`insert_batch()` optimization** — single sealed check + arena lookup for entire batch
-- **Fat LTO + strip** in release profile for maximum codegen quality
+### v0.1.1 Optimizations
+
+- **27% faster concurrent inserts** (4 threads: 14.5ms → 10.6ms)
+- **20% faster sequential inserts** (10K: 1.7ms → 1.36ms)
+- **21% faster random inserts** (10K: 2.8ms → 2.2ms)
+- **33% faster cursor seek** (1K: 54ns → 36ns)
+- **19% faster seal+iterate** (1K: 155µs → 126µs)
+- O(1) `memory_usage()` — atomic running total instead of iterating all shards
+- Conditional allocation tracking — zero overhead when memory limits aren't configured
+- Bulk u64 header writes in `init_node()` (3 stores vs 8 individual stores)
+- Adaptive backoff on CAS failure reduces cache-line bouncing under contention
+- Fat LTO + strip in release profile for maximum codegen quality
 
 See [USAGE.md](USAGE.md) for complete API reference.
 

README.md

@@ -482,7 +482,17 @@ Arena memory is bulk-allocated in blocks and bulk-reclaimed when the memtable is
 
 ## Changelog
 
-### v0.1.0 (unreleased)
+### v0.1.1 (2026-03-31)
+
+- **Performance**: Up to 27% faster concurrent inserts, 20% faster sequential inserts
+- O(1) `memory_usage()` via atomic running total (was O(N shards))
+- Adaptive backoff on CAS failure to reduce cache-line bouncing
+- Optimized `insert_batch()`: single sealed check, single arena lookup
+- Bulk u64 header writes in `init_node()` (3 stores vs 8)
+- Conditional allocation tracking — zero overhead when memory limits aren't configured
+- Release profile: `lto = "fat"`, `strip = true`
+
+### v0.1.0 (2026-03-26)
 
 - Initial release
 - Lock-free skip list with per-thread arena shards