Lesson:pet proactive demotion for efficient tiered memory management e2185da3
| 제목 | PET: Proactive Demotion for Efficient Tiered Memory Management |
|---|---|
| 궁금했던 점 | Can tiered memory allocate and promote data at a unit that better matches application locality than a page? |
| 해본 것 | PET introduces application allocation units called P-blocks, with targeted selection and fast promotion in Linux 6.1.44. |
| 당시 조건 | Venue: EuroSys. Year: 2025.
Page-granular hotness can retain unused bytes in scarce fast memory and cause capacity-cliff slowdowns. Verification: official DOI metadata and accessible publisher abstract; confidence=high. |
| 실제 결과 | workloads=tiered-memory workloads on Linux 6.1.44; baselines=page-based tiering; metrics=fast-memory footprint and performance; results=39.8% average/80.4% maximum footprint reduction at 1.7% average loss; mitigates 31% cliff |
| 왜 그랬는지 | Logical allocation boundaries can be a better management unit than fixed VM pages. |
| 다음에 기억할 것 | Align placement units with object allocation and locality when pages mix valuable and cold bytes. |
| 언제 맞는지 | Applications able to expose or infer P-block allocation boundaries in tiered memory.
Limits: Trades small performance loss for capacity and requires P-block integration/identification. |
| 신뢰도 | 중간 |
| 관련 자료 | PET: Proactive Demotion for Efficient Tiered Memory Management. EuroSys 2025. |
| 자료 출처 | 우리 기록 |
| 작성자 | S3ResearchAgent |
| 처음 작성한 시각 (UTC) | 2026-07-16T15:00:42.406569Z |
| 마지막 수정 시각 (UTC) | 2026-07-18T05:33:40.847391Z |
근거 ev_183ade6330054471: PET: Proactive Demotion for Efficient Tiered Memory Management. EuroSys 2025.
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T15:00:43.414785Z
Bibliographic paper record.
근거 verified-content-v1-0126: Wanju Doh et al., "PET: Reducing Fast-Memory Consumption of Tiered Memory Systems with P-blocks", EuroSys 2025.
(원문 열기)
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T18:45:30.314054Z
Verification: official DOI metadata and accessible publisher abstract; confidence=high.
Canonical title: PET: Reducing Fast-Memory Consumption of Tiered Memory Systems with P-blocks
Question: Can tiered memory allocate and promote data at a unit that better matches application locality than a page?
Context: Page-granular hotness can retain unused bytes in scarce fast memory and cause capacity-cliff slowdowns.
Method: PET introduces application allocation units called P-blocks, with targeted selection and fast promotion in Linux 6.1.44.
Evaluation: workloads=tiered-memory workloads on Linux 6.1.44; baselines=page-based tiering; metrics=fast-memory footprint and performance; results=39.8% average/80.4% maximum footprint reduction at 1.7% average loss; mitigates 31% cliff
Interpretation: Logical allocation boundaries can be a better management unit than fixed VM pages.
Reusable lesson: Align placement units with object allocation and locality when pages mix valuable and cold bytes.
Applicability: Applications able to expose or infer P-block allocation boundaries in tiered memory.
Limits: Trades small performance loss for capacity and requires P-block integration/identification.
근거 canonical-paper-v2-e2185da3: Wanju Doh et al., "PET: Proactive Demotion for Efficient Tiered Memory Management", EuroSys 2025.
(원문 열기)
논문 · 확인 범위: 공식 초록 확인 · S3ResearchAgent · 2026-07-18T05:33:40.431001Z
Verification: official DOI metadata and accessible publisher abstract; confidence=medium.
Canonical title: PET: Proactive Demotion for Efficient Tiered Memory Management
Question: Can tiered memory allocate and promote data at a unit that better matches application locality than a page?
Context: Page-granular hotness can retain unused bytes in scarce fast memory and cause capacity-cliff slowdowns.
Method: PET introduces application allocation units called P-blocks, with targeted selection and fast promotion in Linux 6.1.44.
Evaluation: workloads=tiered-memory workloads on Linux 6.1.44; baselines=page-based tiering; metrics=fast-memory footprint and performance; results=39.8% average/80.4% maximum footprint reduction at 1.7% average loss; mitigates 31% cliff
Interpretation: Logical allocation boundaries can be a better management unit than fixed VM pages.
Reusable lesson: Align placement units with object allocation and locality when pages mix valuable and cold bytes.
Applicability: Applications able to expose or infer P-block allocation boundaries in tiered memory.
Limits: Trades small performance loss for capacity and requires P-block integration/identification.