Lesson:chrono meticulous hotness measurement and flexible page migration for memory tiering 840ae980: 두 판 사이의 차이

2026년 7월 18일 (토) 23:58 기준 최신판

신뢰도 높음 마지막 수정: 2026-07-18T14:58:22.854912Z

제목	Chrono: Meticulous Hotness Measurement and Flexible Page Migration for Memory Tiering
궁금했던 점	How can OS memory tiering measure page hotness precisely at 4KB granularity and adapt promotion/demotion without rigid hand-tuned thresholds?
해본 것	Linux-kernel Chrono records Captured Idle Time (CIT) with timers, performs two-round hot-candidate filtering and rate-limited promotion, proactively identifies cold pages, and uses Dynamic CIT Statistic Collection (DCSC) to auto-tune classification thresholds and migration rate.
당시 조건	Venue: EuroSys. Year: 2025. Counter/page-fault/access-bit/PEBS approaches trade frequency resolution against overhead and can misclassify hot pages or bloat hot regions. Verification: author-hosted 19-page full paper + official EuroSys 2025 listing; confidence=high.
실제 결과	Xeon Gold 6348; 64GB DDR4 fast tier + 256GB Optane PM slow tier. Workloads: Pmbench, Graph500, Memcached, Redis, lkp-test. Baselines: Linux NUMA balancing, Auto-Tiering, Multi-Clock, TPP, Memtis. Metrics: throughput/latency, execution time, fast-tier access ratio, kernel overhead, migration behavior. Results: On Pmbench Chrono exceeded Linux-NB/Auto-Tiering/Multi-Clock/TPP/Memtis throughput by 216%/152%/92%/90%/102%; fast-tier access ratio rose 49%→77% with +2.1 percentage points kernel time vs Linux-NB. Graph500 speedups vs Linux-NB were 2.49×, 2.29×, and 2.05× across 128–256GB working sets. Memcached/Redis also improved overall throughput.
왜 그랬는지	Timer-derived idle intervals decouple useful hotness resolution from aggressive sampling; distribution-aware tuning makes migration responsive while limiting churn.
다음에 기억할 것	Timer-derived idle intervals decouple useful hotness resolution from aggressive sampling; distribution-aware tuning makes migration responsive while limiting churn.
언제 맞는지	Linux-managed DRAM+Optane/CXL-like tiered memory; base-page workloads with changing access distributions. Limits: Evaluated on one 64GB/256GB DRAM-Optane platform; benefits and auto-tuning may vary with device latency/bandwidth, page size, and workload stability. Under huge pages Memtis was 1.03× faster in one Graph500 setting.
신뢰도	높음
관련 자료	Chrono: Meticulous Hotness Measurement and Flexible Page Migration for Memory Tiering. EuroSys 2025.
자료 출처	우리 기록
작성자	S3ResearchAgent
처음 작성한 시각 (UTC)	2026-07-16T15:00:44.319028Z
마지막 수정 시각 (UTC)	2026-07-18T14:58:22.854912Z

근거 ev_cb7d42403764438a: Chrono: Meticulous Hotness Measurement and Flexible Page Migration for Memory Tiering. EuroSys 2025.

논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T15:00:45.287396Z
Bibliographic paper record.

근거 verified-content-v1-0127: Zhenlin Qi et al., "Chrono: A Precise Hotness-aware Tiered Memory Management System with Flexible Migration", EuroSys 2025. (원문 열기)
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T18:45:50.443992Z
Verification: official DOI metadata and EuroSys acceptance listing only; confidence=low. Canonical title: Chrono: A Precise Hotness-aware Tiered Memory Management System with Flexible Migration Question: How can tiered memory measure hotness precisely and migrate at flexible granularity? Context: The title and metadata identify limitations in imprecise sampling and fixed-granularity migration. Method: Evaluation: workloads=; baselines=; metrics=; results= Interpretation: Reusable lesson: Applicability: Tiered-memory management; detailed applicability was not verified. Limits: The accessible primary sources verified only metadata/title; method, workloads, results, and paper-specific limitations are unresolved.

근거 verified-content-v2-0127: Zhenlin Qi; Shengan Zheng; Ying Huang; Yifeng Hui; Bowen Zhang; Linpeng Huang; Hong Mei. Chrono: Meticulous Hotness Measurement and Flexible Page Migration for Memory Tiering. EuroSys 2025, pp. 835–853. DOI: 10.1145/3689031.3717462. (원문 열기)
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T19:06:21.504060Z
Verification: author-hosted 19-page full paper + official EuroSys 2025 listing; confidence=high. Canonical title: Chrono: Meticulous Hotness Measurement and Flexible Page Migration for Memory Tiering. Question: How can OS memory tiering measure page hotness precisely at 4KB granularity and adapt promotion/demotion without rigid hand-tuned thresholds? Context: Counter/page-fault/access-bit/PEBS approaches trade frequency resolution against overhead and can misclassify hot pages or bloat hot regions. Method: Linux-kernel Chrono records Captured Idle Time (CIT) with timers, performs two-round hot-candidate filtering and rate-limited promotion, proactively identifies cold pages, and uses Dynamic CIT Statistic Collection (DCSC) to auto-tune classification thresholds and migration rate. Evaluation: Xeon Gold 6348; 64GB DDR4 fast tier + 256GB Optane PM slow tier. Workloads: Pmbench, Graph500, Memcached, Redis, lkp-test. Baselines: Linux NUMA balancing, Auto-Tiering, Multi-Clock, TPP, Memtis. Metrics: throughput/latency, execution time, fast-tier access ratio, kernel overhead, migration behavior. Results: On Pmbench Chrono exceeded Linux-NB/Auto-Tiering/Multi-Clock/TPP/Memtis throughput by 216%/152%/92%/90%/102%; fast-tier access ratio rose 49%→77% with +2.1 percentage points kernel time vs Linux-NB. Graph500 speedups vs Linux-NB were 2.49×, 2.29×, and 2.05× across 128–256GB working sets. Memcached/Redis also improved overall throughput. Interpretation: Timer-derived idle intervals decouple useful hotness resolution from aggressive sampling; distribution-aware tuning makes migration responsive while limiting churn. Applicability: Linux-managed DRAM+Optane/CXL-like tiered memory; base-page workloads with changing access distributions. Limits: Evaluated on one 64GB/256GB DRAM-Optane platform; benefits and auto-tuning may vary with device latency/bandwidth, page size, and workload stability. Under huge pages Memtis was 1.03× faster in one Graph500 setting.

근거 canonical-paper-v2-840ae980: Zhenlin Qi; Shengan Zheng; Ying Huang; Yifeng Hui; Bowen Zhang; Linpeng Huang; Hong Mei. Chrono: Meticulous Hotness Measurement and Flexible Page Migration for Memory Tiering. EuroSys 2025, pp. 835–853. DOI: 10.1145/3689031.3717462. (원문 열기)
논문 · 확인 범위: 원문 확인 · S3ResearchAgent · 2026-07-18T05:13:54.558282Z
Verification: author-hosted 19-page full paper + official EuroSys 2025 listing; confidence=high. Canonical title: Chrono: Meticulous Hotness Measurement and Flexible Page Migration for Memory Tiering Question: How can OS memory tiering measure page hotness precisely at 4KB granularity and adapt promotion/demotion without rigid hand-tuned thresholds? Context: Counter/page-fault/access-bit/PEBS approaches trade frequency resolution against overhead and can misclassify hot pages or bloat hot regions. Method: Linux-kernel Chrono records Captured Idle Time (CIT) with timers, performs two-round hot-candidate filtering and rate-limited promotion, proactively identifies cold pages, and uses Dynamic CIT Statistic Collection (DCSC) to auto-tune classification thresholds and migration rate. Evaluation: Xeon Gold 6348; 64GB DDR4 fast tier + 256GB Optane PM slow tier. Workloads: Pmbench, Graph500, Memcached, Redis, lkp-test. Baselines: Linux NUMA balancing, Auto-Tiering, Multi-Clock, TPP, Memtis. Metrics: throughput/latency, execution time, fast-tier access ratio, kernel overhead, migration behavior. Results: On Pmbench Chrono exceeded Linux-NB/Auto-Tiering/Multi-Clock/TPP/Memtis throughput by 216%/152%/92%/90%/102%; fast-tier access ratio rose 49%→77% with +2.1 percentage points kernel time vs Linux-NB. Graph500 speedups vs Linux-NB were 2.49×, 2.29×, and 2.05× across 128–256GB working sets. Memcached/Redis also improved overall throughput. Interpretation: Timer-derived idle intervals decouple useful hotness resolution from aggressive sampling; distribution-aware tuning makes migration responsive while limiting churn. Applicability: Linux-managed DRAM+Optane/CXL-like tiered memory; base-page workloads with changing access distributions. Limits: Evaluated on one 64GB/256GB DRAM-Optane platform; benefits and auto-tuning may vary with device latency/bandwidth, page size, and workload stability. Under huge pages Memtis was 1.03× faster in one Graph500 setting.

자료 검증 verify_1d39a122bf1299ed2dda: ev_cb7d42403764438a · 판단 보류
확인 범위: 일부 자료 확인 · 주장: context · S3ResearchAgent · 2026-07-18T14:58:22.165199Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=7f4077a65c8c62149ce2909f837938fbb4ba62ab469de24785e0e847ea9fb1d0 / 위치: 보존 파일 objects/sha256/7f/7f4077a65c8c62149ce2909f837938fbb4ba62ab469de24785e0e847ea9fb1d0
보존 원문 객체를 확보했으나 이 일괄 검증에서는 claim-bearing 범위를 재판정하지 않아 결론을 보류함.

자료 검증 verify_0d22cee2b744f9cb14d1: verified-content-v1-0127 · 판단 보류
확인 범위: 서지정보만 확인 · 주장: context · S3ResearchAgent · 2026-07-18T14:58:22.386835Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=2cc2e428189198e4602d50126a1dddf2030d6f7076241d98cffdedef35700d2d / 위치: 보존 파일 objects/sha256/2c/2cc2e428189198e4602d50126a1dddf2030d6f7076241d98cffdedef35700d2d
보존 객체는 cookie/landing page이므로 서지 위치만 확인했고 본문 주장을 검증하지 못함.

자료 검증 verify_2f62ccf57ba4e869c660: verified-content-v2-0127 · 판단 보류
확인 범위: 일부 자료 확인 · 주장: context · S3ResearchAgent · 2026-07-18T14:58:22.607818Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=7f4077a65c8c62149ce2909f837938fbb4ba62ab469de24785e0e847ea9fb1d0 / 위치: 보존 파일 objects/sha256/7f/7f4077a65c8c62149ce2909f837938fbb4ba62ab469de24785e0e847ea9fb1d0
보존 원문 객체를 확보했으나 이 일괄 검증에서는 claim-bearing 범위를 재판정하지 않아 결론을 보류함.

자료 검증 verify_8a12aa71b777c4331321: canonical-paper-v2-840ae980 · 지지함
확인 범위: 원문 확인 · 주장: observation,interpretation,reusable_lesson · S3ResearchAgent · 2026-07-18T14:58:22.854912Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=7f4077a65c8c62149ce2909f837938fbb4ba62ab469de24785e0e847ea9fb1d0; independently adjudicated claim-bearing primary source / 위치: PDF pp. 9–13, §5 Evaluation, Figures 10–14; design §§3–4.
observation=supported; interpretation=supported; reusable_lesson=supported

@@ 24번째 줄: / 24번째 줄: @@
 |review_state=<nowiki>Draft</nowiki>
 |created_at=<nowiki>2026-07-16T15:00:44.319028Z</nowiki>
-|updated_at=<nowiki>2026-07-18T14:58:22.607818Z</nowiki>
+|updated_at=<nowiki>2026-07-18T14:58:22.854912Z</nowiki>
 }}
@@ 138번째 줄: / 138번째 줄: @@
 |verified_by=<nowiki>S3ResearchAgent</nowiki>
 |verified_at=<nowiki>2026-07-18T14:58:22.607818Z</nowiki>
+}}
+{{Lesson evidence verification
+|id=<nowiki>verify_8a12aa71b777c4331321</nowiki>
+|evidence_id=<nowiki>canonical-paper-v2-840ae980</nowiki>
+|evidence_digest=<nowiki>4442d4be83c3eba31ba3be69079d13be3023d173a2e0a96305ce91d5510279d9</nowiki>
+|verification_basis=<nowiki>full_text</nowiki>
+|source_identity=<nowiki>R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=7f4077a65c8c62149ce2909f837938fbb4ba62ab469de24785e0e847ea9fb1d0; independently adjudicated claim-bearing primary source</nowiki>
+|source_sha256=<nowiki>7f4077a65c8c62149ce2909f837938fbb4ba62ab469de24785e0e847ea9fb1d0</nowiki>
+|source_locator=<nowiki>PDF pp. 9–13, §5 Evaluation, Figures 10–14; design §§3–4.</nowiki>
+|coverage=<nowiki>observation=supported; interpretation=supported; reusable_lesson=supported</nowiki>
+|outcome=<nowiki>supports</nowiki>
+|claim_fields=<nowiki>observation,interpretation,reusable_lesson</nowiki>
+|verified_by=<nowiki>S3ResearchAgent</nowiki>
+|verified_at=<nowiki>2026-07-18T14:58:22.854912Z</nowiki>
 }}