속성:Evidence note
자료에서 확인할 수 있는 점이나 한계를 적습니다.
m
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1. +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_managing_gpu_buffers_for_caching_more_apps_in_mobile_systems_46ece2ab. 확인 범위: official_abstract. 확인한 자료: https://doi.org/10.1109/EMSOFT.2015.7318276 ; https://www.sigbed.org/emsoft-info/participants/Jeong%2C%20Jinkyu.html ; https://yonsei.elsevierpure.com/en/publications/managing-gpu-buffers-for-caching-more-apps-in-mobile-systems/. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다. +
Verification: full_text; confidence=high.
Question: Can virtualized clouds use CXL-attached memory without software-tiering overhead or application-oblivious hardware slowdowns?
Context: Intel Flat Memory Mode (FMM) is the first hardware-managed CXL tiering mechanism, but workload sensitivity creates large outliers.
Method: Memstrata combines FMM with a page-coloring allocator and an online estimator that controls each VM's fast-tier share by measured slowdown.
Evaluation: workloads=full-system multi-VM workloads; baselines=Intel FMM; software tiering; metrics=performance degradation; slowdown outliers; results=FMM <=5% degradation for >82% of workloads; worst FMM outlier 34%; Memstrata cut >30% degradation to <6%
Interpretation: Hardware tiering is broadly useful, but per-VM allocation and feedback are necessary to control tail slowdowns.
Reusable lesson: Pair transparent hardware tiering with isolation-aware allocation and online slowdown estimation.
Applicability: Virtualized CXL memory pools and multi-tenant cloud hosts.
Limits: Results depend on the evaluated FMM/CXL prototype, workload mix, and estimator/page-coloring assumptions. +
Bibliographic paper record. +
Verification: full_text; confidence=high.
Question: Can virtualized clouds use CXL-attached memory without software-tiering overhead or application-oblivious hardware slowdowns?
Context: Intel Flat Memory Mode (FMM) is the first hardware-managed CXL tiering mechanism, but workload sensitivity creates large outliers.
Method: Memstrata combines FMM with a page-coloring allocator and an online estimator that controls each VM's fast-tier share by measured slowdown.
Evaluation: workloads=full-system multi-VM workloads; baselines=Intel FMM; software tiering; metrics=performance degradation; slowdown outliers; results=FMM <=5% degradation for >82% of workloads; worst FMM outlier 34%; Memstrata cut >30% degradation to <6%
Interpretation: Hardware tiering is broadly useful, but per-VM allocation and feedback are necessary to control tail slowdowns.
Reusable lesson: Pair transparent hardware tiering with isolation-aware allocation and online slowdown estimation.
Applicability: Virtualized CXL memory pools and multi-tenant cloud hosts.
Limits: Results depend on the evaluated FMM/CXL prototype, workload mix, and estimator/page-coloring assumptions. +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1. +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_memory_deduplication_in_mobile_systems_9c6aec68. 확인 범위: metadata_only. 확인한 자료: https://casyslab.kaist.ac.kr/publications/international-conference ; https://cslab.yonsei.ac.kr/publications. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다. +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_memsos_os_guided_selective_memory_mirroring_4aec983d. 확인 범위: official_abstract. 확인한 자료: https://2026.hpca-conf.org/details/hpca-2026-main-conference/20/MemSOS-OS-Guided-Selective-Memory-Mirroring ; https://doi.org/10.1109/HPCA68181.2026.11408560. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다. +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1. +
midas minimizing write amplification in log structured systems through adaptive group number and 1cac80b9 +
Bibliographic paper record. +
midas minimizing write amplification in log structured systems through adaptive group number and 1cac80b9 +
Verification: abstract_only; confidence=high.
Canonical title: MIDAS: Minimizing Write Amplification in Log-Structured Systems through Adaptive Group Number and Size Configuration
Question: 로그 구조 저장소의 그룹 수·크기를 워크로드에 맞춰 자동 조정해 write amplification을 줄일 수 있는가?
Context: 고정 hot/cold 그룹은 데이터 수명 분포 변화에 따라 GC 복사를 늘린다.
Method: age 기반 chain group, UID+Markov 모델, 동적 HOT group으로 그룹 수·크기를 조정한다.
Evaluation: workloads=simulation; flash SSD proof-of-concept; baselines=state-of-the-art garbage collection; metrics=write amplification; throughput; CPU/memory overhead; results=25% lower WAF; 54% higher throughput.
Interpretation: 데이터 수명 분포를 온라인 추정해 GC 격리 단위를 바꾸면 복사를 줄일 수 있다.
Reusable lesson: log-structured GC의 그룹 구성은 고정 상수가 아니라 적응 변수다.
Applicability: LSM·로그 구조 SSD/스토리지.
Limits: 트레이스 정상성·Markov 예측과 프로토타입 SSD 구성에 의존한다. +
midas minimizing write amplification in log structured systems through adaptive group number and 1cac80b9 +
Verification: abstract_only; confidence=medium.
Canonical title: MIDAS: Minimizing Write Amplification in Log-Structured Systems through Adaptive Group Number and Size Configuration
Question: 로그 구조 저장소의 그룹 수·크기를 워크로드에 맞춰 자동 조정해 write amplification을 줄일 수 있는가?
Context: 고정 hot/cold 그룹은 데이터 수명 분포 변화에 따라 GC 복사를 늘린다.
Method: age 기반 chain group, UID+Markov 모델, 동적 HOT group으로 그룹 수·크기를 조정한다.
Evaluation: workloads=simulation; flash SSD proof-of-concept; baselines=state-of-the-art garbage collection; metrics=write amplification; throughput; CPU/memory overhead; results=25% lower WAF; 54% higher throughput.
Interpretation: 데이터 수명 분포를 온라인 추정해 GC 격리 단위를 바꾸면 복사를 줄일 수 있다.
Reusable lesson: log-structured GC의 그룹 구성은 고정 상수가 아니라 적응 변수다.
Applicability: LSM·로그 구조 SSD/스토리지.
Limits: 트레이스 정상성·Markov 예측과 프로토타입 SSD 구성에 의존한다. +
minflow high performance and cost efficient data passing for i o intensive stateful serverless a 886ce161 +
Verification: abstract_only; confidence=high.
Canonical title: MinFlow: High-performance and Cost-efficient Data Passing for I/O-intensive Stateful Serverless Analytics
Question: I/O 집약 서버리스 분석의 함수 간 데이터 전달을 빠르고 싸게 만들 수 있는가?
Context: 원격 저장소를 거치는 DAG shuffle이 지연과 비용을 지배한다.
Method: PUT/GET을 줄이는 다단 토폴로지, interleaved partition, 모델 기반 구성 선택을 결합한다.
Evaluation: workloads=I/O-intensive serverless analytics; baselines=FaaSFlow; Lambada; metrics=job completion time; storage cost; remote-storage traffic; results=>50% remote-storage traffic eliminated; exact JCT/cost aggregate not abstract-verified.
Interpretation: 워크플로 topology 자체를 데이터 전달 비용에 맞춰 생성해야 한다.
Reusable lesson: serverless shuffle은 전송 API보다 DAG·partition 공동 설계가 중요하다.
Applicability: stateful serverless analytics·shuffle.
Limits: 워크플로 구조·성능 모델·클라우드 가격의 정확성에 의존한다. +
minflow high performance and cost efficient data passing for i o intensive stateful serverless a 886ce161 +
Verification: abstract_only; confidence=medium.
Canonical title: MinFlow: High-performance and Cost-efficient Data Passing for I/O-intensive Stateful Serverless Analytics
Question: I/O 집약 서버리스 분석의 함수 간 데이터 전달을 빠르고 싸게 만들 수 있는가?
Context: 원격 저장소를 거치는 DAG shuffle이 지연과 비용을 지배한다.
Method: PUT/GET을 줄이는 다단 토폴로지, interleaved partition, 모델 기반 구성 선택을 결합한다.
Evaluation: workloads=I/O-intensive serverless analytics; baselines=FaaSFlow; Lambada; metrics=job completion time; storage cost; remote-storage traffic; results=>50% remote-storage traffic eliminated; exact JCT/cost aggregate not abstract-verified.
Interpretation: 워크플로 topology 자체를 데이터 전달 비용에 맞춰 생성해야 한다.
Reusable lesson: serverless shuffle은 전송 API보다 DAG·partition 공동 설계가 중요하다.
Applicability: stateful serverless analytics·shuffle.
Limits: 워크플로 구조·성능 모델·클라우드 가격의 정확성에 의존한다. +
minflow high performance and cost efficient data passing for i o intensive stateful serverless a 886ce161 +
Bibliographic paper record. +
Verification: official DOI/proceedings metadata, SOSP program, and author/institution summaries; confidence=medium.
Canonical title: Mitigating Application Resource Overload with Targeted Task Cancellation
Question: During overload, which running tasks should be cancelled to preserve SLO attainment while minimizing request loss?
Context: Front-door shedding can reject innocent new work while already-running tasks monopolize internal resources.
Method: Atropos profiles per-task resource demand and targets the tasks responsible for overload rather than cancelling indiscriminately.
Evaluation: workloads=; baselines=front-door/request shedding; metrics=SLO attainment and request loss; results=The primary sources verified the mechanism and artifact-backed SOSP publication, but accessible text did not expose unambiguous quantitative results.
Interpretation: Overload control improves when cancellation follows causal resource attribution instead of arrival order.
Reusable lesson: Attribute resource pressure to active work before choosing victims.
Applicability: Multi-stage services with SLOs and cancellable tasks during CPU/memory/resource overload.
Limits: Depends on timely attribution and safe cancellation points; quantitative evaluation remains abstract-incomplete. +
Bibliographic paper record. +
Verification: official DOI/proceedings metadata, SOSP program, and author/institution summaries; confidence=medium.
Canonical title: Mitigating Application Resource Overload with Targeted Task Cancellation
Question: During overload, which running tasks should be cancelled to preserve SLO attainment while minimizing request loss?
Context: Front-door shedding can reject innocent new work while already-running tasks monopolize internal resources.
Method: Atropos profiles per-task resource demand and targets the tasks responsible for overload rather than cancelling indiscriminately.
Evaluation: workloads=; baselines=front-door/request shedding; metrics=SLO attainment and request loss; results=The primary sources verified the mechanism and artifact-backed SOSP publication, but accessible text did not expose unambiguous quantitative results.
Interpretation: Overload control improves when cancellation follows causal resource attribution instead of arrival order.
Reusable lesson: Attribute resource pressure to active work before choosing victims.
Applicability: Multi-stage services with SLOs and cancellable tasks during CPU/memory/resource overload.
Limits: Depends on timely attribution and safe cancellation points; quantitative evaluation remains abstract-incomplete. +
mitigating resource usage dependency in sorting based kv stores on hybrid storage devices via op 31cbb8ad +
Verification: official USENIX page and abstract; confidence=high.
Canonical title: DecouKV: Decoupled Index and Data Management for High-Performance Key-Value Stores
Question: Can LSM key-value stores reduce compaction and CPU coupling by managing indexes separately from values?
Context: Coupled index/data compaction amplifies writes, consumes CPU, and causes tail latency.
Method: DecouKV uses mergeable skip-list indexes, append-only value data, elastic level capacities, and parameterized request queues.
Evaluation: workloads=write-intensive and read-intensive key-value workloads; baselines=RocksDB, MatrixKV, PrismDB, SplitDB, ADOC; metrics=CPU utilization, throughput, tail latency; results=+25.4–32.3% CPU utilization; 2.3–4.9x write throughput; 74.3–91.4% lower tail; 1.2–2.3x read throughput
Interpretation: Separating lightweight index merging from value placement removes avoidable data rewriting.
Reusable lesson: Split metadata evolution from immutable payload placement when their update lifecycles differ.
Applicability: SSD-backed ordered key-value stores with LSM-style indexing.
Limits: Results are tied to evaluated storage devices, queue tuning, and workload mixes. +
mitigating resource usage dependency in sorting based kv stores on hybrid storage devices via op 31cbb8ad +
Bibliographic paper record. +