Files
hakmem/docs/paper/ACE-Alloc/main.md

113 lines
7.9 KiB
Markdown
Raw Normal View History

% ACEAlloc: Agentic Context Engineering for RuntimeAdaptive SmallObject Allocation
% Authors: (TBD)
% Draft v0.1 — 20251102
概要Abstract
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
本論文は、Agentic Context EngineeringACEをメモリアロケータに適用し、Box Theory に基づく TwoSpeed Tiny フロントHOT/WARM/COLDと低オーバーヘッドの学習層を備えた小型オブジェクト向けアロケータ「ACEAlloc」を提案する。ACEAlloc は、観測軽量イベント、意思決定cap/refill/trim の動的制御)、適用(非同期スレッド)から成るエージェント型最適化ループを実装しつつ、ホットパスから観測負荷を排除する TLS バッチ化を採用する。また、標準 API 互換の free(ptr) を保ちながら perobject ヘッダを削除し、スラブ末尾の 32B prefix メタデータと Tiny Front Gatekeeper/Route Box により密度劣化なく即時判定を実現する。Tinyonly のホットパスベンチマークでは mimalloc と同一オーダーのスループットを達成しつつ、Mixed/Tiny+Mid のワークロードでは Refillone、SLL 縮小、Idle Trim、および Superslab Tiering の ACE 制御により性能とメモリ効率のトレードオフを系統的に探索可能であることを示す。
1. はじめにIntroduction
- 小型オブジェクトの高速割り当ては、多くのアプリケーションにとって主要な性能決定因子である。
- 既存の高速アロケータは、性能とメモリ効率のトレードオフに直面する。エージェント型の制御ACEにより、ワークロードに追従した動的最適化を可能にする。
- 本論文の貢献:
- ACEAgentic Context Engineeringをメモリアロケータに実装し、低オーバーヘッドで動作させる設計TLS バッチ化、非同期適用)。
- 標準 API に準拠しながら perobject ヘッダを排し、スラブ末尾 32B prefix で即時判定を実現する設計。
- 性能/メモリのスイートスポット探索Refillone、SLL 縮小、Idle Trim/Flushと評価。
2. 背景Background
- メモリアロケータの基礎小型クラス、TLS キャッシュ、スラブ/スーパー・スラブ)。
- mimalloc などの関連実装の要点(ページ記述子、低メタデータ、局所性)。
- Agentic Context EngineeringACEの要旨観測→意思決定→適用のエージェント型ループを、ホットパスを汚染せず実装する考え方。
3. 設計Design of ACEAlloc
- 目標:
- ホットパスの命令・分岐・メモリアクセスを最小化(ゼロに近い)。
- 標準 API 互換free(ptr))とメモリ密度の維持。
- 学習層は非同期・オフホットパスで適用。
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- Box Theory に基づき、ホットパスTiny Frontと学習層ACE/ELO/CAPを明確に分離した TwoSpeed 構成とする。
- キー設計:
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- TwoSpeed Tiny Front: HOT パスTLS SLL / Unified Cache、WARM パスバッチリフィル、COLD パスShared Pool / Superslab / Registryを箱として分離し、HOT パスから Registry 参照・mutex・重い診断を排除する。
- TLS バッチ化alloc/free の観測カウンタは TLS に蓄積、しきい値到達時のみ atomic 反映)。
- 観測リング+背景ワーカー(イベントの集約とポリシ適用)。
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- スラブ末尾 32B prefixpool/type/class/owner を格納)と Tiny Layout/Ptr Bridge Box により perobject ヘッダを不要化。
- Tiny Front Gatekeeper / Route Box により、malloc/free の入口で USER→BASE 変換と Tiny vs Pool のルーティングを 1 箇所に集約。
- Refilloneミス時 1 個だけ補充)と SLL 縮小、Idle Trim/Flush、Superslab TieringHOT/DRAINING/FREEのポリシ。
4. 実装Implementation
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- Tiny / Superslab の Box 化:
- Tiny FrontHOT/WARM/COLD: `core/box/tiny_front_hot_box.h``core/box/tiny_front_cold_box.h``core/box/tiny_alloc_gate_box.h``core/box/tiny_free_gate_box.h``core/box/tiny_route_box.{h,c}`
- Unified Cache / Backend: `core/tiny_unified_cache.{h,c}``core/hakmem_shared_pool_*.c``core/box/ss_allocation_box.{h,c}`
- Superslab Tiering / Release Guard: `core/box/ss_tier_box.h``core/box/ss_release_guard_box.h``core/hakmem_super_registry.{c,h}`
- Headerless + ポインタ変換:
- Prefix メタデータとレイアウト: `core/hakmem_tiny_superslab*.h``core/box/tiny_layout_box.h``core/box/tiny_header_box.h`
- USER/BASE ブリッジ: `core/box/tiny_ptr_bridge_box.h`、TLS SLL / Remote Queue: `core/box/tls_sll_box.h``core/box/tls_sll_drain_box.h`
- 学習層ACE/ELO/CAP
- ACE メトリクスとコントローラ: `core/hakmem_ace_metrics.{h,c}``core/hakmem_ace_controller.{h,c}``core/hakmem_elo.{h,c}``core/hakmem_learner.{h,c}`
- INT エンジン: `core/hakmem_tiny_intel.inc`(観測→意思決定→適用のループ。デフォルトでは OFF または OBSERVE モードで運用)。
- 互換性と安全性:
- 標準 API と LD_PRELOAD 環境での安全モード(外部アプリから free(ptr) をそのまま受け入れる)。
- Tiny Front Gatekeeper Box による free 境界での検証USER→BASE 正規化、範囲チェック、Box 境界での FailFast
- Remote free は専用の Remote Queue Box に隔離し、オーナーシップ移譲と drain/publish/adopt を Box 境界で分離。
5. 評価Evaluation
- ベンチマークTiny Hot、Mid MT、Mixed本リポジトリ同梱
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- Tiny Hot: `bench_tiny_hot_hakmem`(固定サイズ Tiny クラス、TwoSpeed Tiny Front の HOT パス性能を測定)。
- Mixed: `bench_random_mixed_hakmem`(ランダムサイズ + malloc/free 混在、HOT/WARM/COLD パスの比率も観測)。
- 指標スループットM ops/sec、帯域、RSS/VmSize、断片化比オプション
- 比較mimalloc、システム malloc。
- アブレーション:
- ACE OFF 対比(学習層無効)。
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- TwoSpeed Tiny Front の ON/OFFTiny Route Profile による Tinyonly/Tinyfirst/Poolonly の切り替え)。
- Superslab Tiering / Eager FREE の有無。
- Refillone/SLL 縮小/Idle Trim の有無。
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- Prefix メタ(ヘッダ無し) vs perobject ヘッダ(参考、比較実装がある場合)。
6. 関連研究Related Work
- 既存アロケータmimalloc、jemalloc など)。
- 動的最適化・学習型手法(エージェントベースの最適化)。
7. 考察・限界Discussion & Limitations
- Idle Trim の即時 RSS への効き方(短時間測定 vs 長時間)。
- Remote free 多発時の設計トレードオフ(将来の安全なリモートキュー)。
- MT スケール時のスラブ粒度と ACE ポリシのチューニング。
8. 結論Conclusion
- ACEAlloc は、ホットパスのオーバーヘッドを極小化しつつ、学習層によって動的にメモリ効率と性能のスイートスポットを探索できる実装である。今後は、長時間の断片化評価と MT スケール最適化を進める。
付録 A. Artifact再現手順
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- ビルドTiny/Mixed ベンチ):
```sh
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
make bench_tiny_hot_hakmem bench_random_mixed_hakmem
```
- Tiny性能:
```sh
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
HAKMEM_TINY_PROFILE=full ./bench_tiny_hot_hakmem 64 100 60000
```
- Mixed性能:
```sh
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
HAKMEM_TINY_PROFILE=conservative ./bench_random_mixed_hakmem 2000000 400 42
```
- スイープ計測短時間のCSV出力:
```sh
scripts/sweep_mem_perf.sh both | tee sweep.csv
```
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
- INT エンジン+学習層 ON:
```sh
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
HAKMEM_INT_ENGINE=1 \
./bench_random_mixed_hakmem 2000000 400 42 2>&1 | less
```
Implement Warm Pool Secondary Prefill Optimization (Phase B-2c Complete) Problem: Warm pool had 0% hit rate (only 1 hit per 3976 misses) despite being implemented, causing all cache misses to go through expensive superslab_refill registry scans. Root Cause Analysis: - Warm pool was initialized once and pushed a single slab after each refill - When that slab was exhausted, it was discarded (not pushed back) - Next refill would push another single slab, which was immediately exhausted - Pool would oscillate between 0 and 1 items, yielding 0% hit rate Solution: Secondary Prefill on Cache Miss When warm pool becomes empty, we now do multiple superslab_refills and prefill the pool with 3 additional HOT superlslabs before attempting to carve. This builds a working set of slabs that can sustain allocation pressure. Implementation Details: - Modified unified_cache_refill() cold path to detect empty pool - Added prefill loop: when pool count == 0, load 3 extra superlslabs - Store extra slabs in warm pool, keep 1 in TLS for immediate carving - Track prefill events in g_warm_pool_stats[].prefilled counter Results (1M Random Mixed 256B allocations): - Before: C7 hits=1, misses=3976, hit_rate=0.0% - After: C7 hits=3929, misses=3143, hit_rate=55.6% - Throughput: 4.055M ops/s (maintained vs 4.07M baseline) - Stability: Consistent 55.6% hit rate at 5M allocations (4.102M ops/s) Performance Impact: - No regression: throughput remained stable at ~4.1M ops/s - Registry scan avoided in 55.6% of cache misses (significant savings) - Warm pool now functioning as intended with strong locality Configuration: - TINY_WARM_POOL_MAX_PER_CLASS increased from 4 to 16 to support prefill - Prefill budget hardcoded to 3 (tunable via env var if needed later) - All statistics always compiled, ENV-gated printing via HAKMEM_WARM_POOL_STATS=1 Next Steps: - Monitor for further optimization opportunities (prefill budget tuning) - Consider adaptive prefill budget based on class-specific hit rates - Validate at larger allocation counts (10M+ pending registry size fix) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2025-12-04 23:31:54 +09:00
(詳細な環境変数とプロファイルは `docs/specs/ENV_VARS.md` を参照。)
謝辞Acknowledgments
TBD