본문으로 이동

속성:Evidence note

S3 연구 메모리

Text

자료에서 확인할 수 있는 점이나 한계를 적습니다.

( | ) (20 | 50 | 100 | 250 | 500) 보기
이 속성을 사용하는 문서 20개를 보여줍니다.
s
scalio scaling up dpu based jbof key value store with nvme of target offload 8db39764 +
Verification: official USENIX page and abstract; confidence=medium. Canonical title: Scalio: Scaling up DPU-based JBOF Key-value Store with NVMe-oF Target Offload Question: How can NVMe-oF storage targets scale across dense SSD enclosures without overloading host CPUs? Context: Centralized target processing and inconsistent cache states limit throughput in DPU/JBOF deployments. Method: Scalio offloads target functions to DPUs and uses RDMA-based coordination to keep DPU DRAM and SSD cache states linearizable. Evaluation: workloads=high-density NVMe SSD configurations; baselines=conventional NVMe-oF target architecture; metrics=throughput and linearizability; results=up to 3.3x throughput Interpretation: DPU offload only scales when cache coherence and target-state ownership are designed together. Reusable lesson: Place control and data processing near the storage fabric, with an explicit consistency protocol. Applicability: NVMe-oF JBOF systems with programmable DPUs. Limits: Requires DPU-capable infrastructure and targets high-density configurations.  +
scalio scaling up dpu based jbof key value store with nvme of target offload 8db39764 +
Verification: official USENIX page and abstract; confidence=high. Canonical title: Scalio: Scalable I/O Virtualization for High-Density Storage Question: How can NVMe-oF storage targets scale across dense SSD enclosures without overloading host CPUs? Context: Centralized target processing and inconsistent cache states limit throughput in DPU/JBOF deployments. Method: Scalio offloads target functions to DPUs and uses RDMA-based coordination to keep DPU DRAM and SSD cache states linearizable. Evaluation: workloads=high-density NVMe SSD configurations; baselines=conventional NVMe-oF target architecture; metrics=throughput and linearizability; results=up to 3.3x throughput Interpretation: DPU offload only scales when cache coherence and target-state ownership are designed together. Reusable lesson: Place control and data processing near the storage fabric, with an explicit consistency protocol. Applicability: NVMe-oF JBOF systems with programmable DPUs. Limits: Requires DPU-capable infrastructure and targets high-density configurations.  +
scheduler support for video oriented multimedia on client side virtualization c8366be4 +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_scheduler_support_for_video_oriented_multimedia_on_client_side_virtualization_69d0c186. 확인 범위: full_text. 확인한 자료: https://doi.org/10.1145/2155555.2155566 ; https://yonsei.elsevierpure.com/en/publications/scheduler-support-for-video-oriented-multimedia-on-client-side-vi/ ; https://www.researchgate.net/publication/221636635_Scheduler_support_for_video-oriented_multimedia_on_client-side_virtualization. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다.  +
scheduler support for video oriented multimedia on client side virtualization c8366be4 +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1.  +
scoz a systemwide causal profiler for multicore systems 7b5e72e2 +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_scoz_a_system_wide_causal_profiler_for_multicore_systems_9ddaf48a. 확인 범위: full_text. 확인한 자료: https://onlinelibrary.wiley.com/doi/full/10.1002/spe.2930 ; https://doi.org/10.1002/spe.2930 ; https://yonsei.elsevierpure.com/en/publications/scoz-a-system-wide-causal-profiler-for-multicore-systems/. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다.  +
scoz a systemwide causal profiler for multicore systems 7b5e72e2 +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1.  +
selective memory deduplication for cost efficiency in mobile smart devices 9f3139a0 +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_selective_memory_deduplication_for_cost_efficiency_in_mobile_smart_devices_ef7d0fd2. 확인 범위: official_abstract. 확인한 자료: https://doi.org/10.1109/TCE.2014.6852004 ; https://yonsei.elsevierpure.com/en/publications/selective-memory-deduplication-for-cost-efficiency-in-mobile-smar/. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다.  +
selective memory deduplication for cost efficiency in mobile smart devices 9f3139a0 +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1.  +
serverless in the wild characterizing and optimizing the serverless workload at a large cloud pr f2fa1121 +
Bibliographic paper record.  +
serverless in the wild characterizing and optimizing the serverless workload at a large cloud pr f2fa1121 +
Verification: abstract_only; confidence=high. Canonical title: Serverless in the Wild: Characterizing and Optimizing the Serverless Workload at a Large Cloud Provider Question: 실제 FaaS 워크로드는 어떤 분포를 보이며 콜드 스타트를 자원 효율적으로 줄일 수 있는가? Context: 공개 실측이 부족해 서버리스 자원 정책이 단순 가정에 의존했다. Method: Azure Functions 14일 플릿을 분석하고 앱별 keep-alive와 ARIMA 기반 예측을 혼합한다. Evaluation: workloads=14-day Azure Functions fleet trace; baselines=fixed keep-alive policies; metrics=cold starts; resource use; results=Significantly fewer cold starts with fewer resources; exact figure not abstract-verified. Interpretation: 애플리케이션별 주기성과 희소성을 분리해 정책을 선택해야 한다. Reusable lesson: 단일 고정 keep-alive보다 워크로드별 하이브리드 정책이 낫다. Applicability: FaaS 용량 계획과 콜드 스타트 완화. Limits: 단일 공급자·14일 관측이며 초록에서 정확한 개선 수치는 확인되지 않는다.  +
serverless in the wild characterizing and optimizing the serverless workload at a large cloud pr f2fa1121 +
Verification: abstract_only; confidence=medium. Canonical title: Serverless in the Wild: Characterizing and Optimizing the Serverless Workload at a Large Cloud Provider Question: 실제 FaaS 워크로드는 어떤 분포를 보이며 콜드 스타트를 자원 효율적으로 줄일 수 있는가? Context: 공개 실측이 부족해 서버리스 자원 정책이 단순 가정에 의존했다. Method: Azure Functions 14일 플릿을 분석하고 앱별 keep-alive와 ARIMA 기반 예측을 혼합한다. Evaluation: workloads=14-day Azure Functions fleet trace; baselines=fixed keep-alive policies; metrics=cold starts; resource use; results=Significantly fewer cold starts with fewer resources; exact figure not abstract-verified. Interpretation: 애플리케이션별 주기성과 희소성을 분리해 정책을 선택해야 한다. Reusable lesson: 단일 고정 keep-alive보다 워크로드별 하이브리드 정책이 낫다. Applicability: FaaS 용량 계획과 콜드 스타트 완화. Limits: 단일 공급자·14일 관측이며 초록에서 정확한 개선 수치는 확인되지 않는다.  +
serverlessllm low latency serverless inference for large language models 69b8cd6d +
Bibliographic paper record.  +
serverlessllm low latency serverless inference for large language models 69b8cd6d +
Verification: full_text; confidence=high. Canonical title: ServerlessLLM: Low-Latency Serverless Inference for Large Language Models Question: How can serverless LLM inference start large models quickly despite GPU-memory scarcity and bursty arrivals? Context: Loading model checkpoints from remote storage dominates cold starts, while keeping every model resident is too costly. Method: ServerlessLLM uses a loading-optimized checkpoint format, multi-tier local loading, live migration, and a scheduler aware of locality and startup cost. Evaluation: workloads=multiple LLM workloads; microbenchmarks; real serverless scenarios; baselines=state-of-the-art serverless inference systems; metrics=startup latency; inference latency; loading throughput; results=10–200x lower latency Interpretation: Model loading must be treated as a scheduled data-placement pipeline rather than a one-off cold-start operation. Reusable lesson: Co-design persistent layout, cache hierarchy, migration, and placement policy for large cold state. Applicability: Multi-model serverless LLM inference clusters. Limits: Benefits depend on local checkpoint capacity, locality, storage bandwidth, model size, and loading-dominated workloads.  +
serverlessllm low latency serverless inference for large language models 69b8cd6d +
Verification: full_text; confidence=high. Canonical title: ServerlessLLM: Low-Latency Serverless Inference for Large Language Models Question: How can serverless LLM inference start large models quickly despite GPU-memory scarcity and bursty arrivals? Context: Loading model checkpoints from remote storage dominates cold starts, while keeping every model resident is too costly. Method: ServerlessLLM uses a loading-optimized checkpoint format, multi-tier local loading, live migration, and a scheduler aware of locality and startup cost. Evaluation: workloads=multiple LLM workloads; microbenchmarks; real serverless scenarios; baselines=state-of-the-art serverless inference systems; metrics=startup latency; inference latency; loading throughput; results=10–200x lower latency Interpretation: Model loading must be treated as a scheduled data-placement pipeline rather than a one-off cold-start operation. Reusable lesson: Co-design persistent layout, cache hierarchy, migration, and placement policy for large cold state. Applicability: Multi-model serverless LLM inference clusters. Limits: Benefits depend on local checkpoint capacity, locality, storage bandwidth, model size, and loading-dominated workloads.  +
shape adaptive attention negative 20260722 +
Remote branch evidence for the code and research notes; the uncommitted L40S graph artifact is bound separately by SHA-256 in the initial benchmark evidence.  +
shiftlock mitigate one sided rdma lock contention via handover a5d67e63 +
Verification: official_abstract; confidence=high. Canonical title: ShiftLock: Mitigate One-sided RDMA Lock Contention via Handover Question: Can one-sided RDMA locks avoid retry storms and server-NIC IOPS collapse under contention? Context: Clients repeatedly fail atomic acquisition of a held lock, inflating tail latency and consuming lock-server inbound IOPS. Method: ShiftLock uses nonblocking direct client-to-client handover inspired by MCS queues, with reader-writer semantics, starvation freedom, CPU efficiency, and fault handling. Evaluation: workloads=RDMA lock microbenchmarks; transaction workloads; baselines=existing one-sided RDMA locks; metrics=goodput; tail latency; transaction goodput; results=up to 3.62x microbenchmark goodput; up to 76.6% lower tail latency; up to 2.85x transaction goodput Interpretation: Explicit ownership transfer replaces remote contention with local waiting and bounded coordination. Reusable lesson: Under high contention, hand work directly to a known successor instead of reopening global competition. Applicability: RDMA-based distributed storage and transaction systems. Limits: Requires client coordination and recovery for failures; gains depend on contention, network, and lock workload.  +
shiftlock mitigate one sided rdma lock contention via handover a5d67e63 +
Verification: official_abstract; confidence=medium. Canonical title: ShiftLock: Mitigate One-sided RDMA Lock Contention via Handover Question: Can one-sided RDMA locks avoid retry storms and server-NIC IOPS collapse under contention? Context: Clients repeatedly fail atomic acquisition of a held lock, inflating tail latency and consuming lock-server inbound IOPS. Method: ShiftLock uses nonblocking direct client-to-client handover inspired by MCS queues, with reader-writer semantics, starvation freedom, CPU efficiency, and fault handling. Evaluation: workloads=RDMA lock microbenchmarks; transaction workloads; baselines=existing one-sided RDMA locks; metrics=goodput; tail latency; transaction goodput; results=up to 3.62x microbenchmark goodput; up to 76.6% lower tail latency; up to 2.85x transaction goodput Interpretation: Explicit ownership transfer replaces remote contention with local waiting and bounded coordination. Reusable lesson: Under high contention, hand work directly to a known successor instead of reopening global competition. Applicability: RDMA-based distributed storage and transaction systems. Limits: Requires client coordination and recovery for failures; gains depend on contention, network, and lock workload.  +
shiftlock mitigate one sided rdma lock contention via handover a5d67e63 +
Bibliographic paper record.  +
sieve is simpler than lru an efficient turn key eviction algorithm for web caches 86f86293 +
Verification: abstract_only; confidence=medium. Canonical title: SIEVE is Simpler than LRU: an Efficient Turn-Key Eviction Algorithm for Web Caches Question: 웹 캐시에 쉽게 이식하면서 LRU보다 단순하고 효율적인 축출 정책을 만들 수 있는가? Context: 최신 정책은 복잡하고 생산 캐시 라이브러리에 넣기 어렵다. Method: FIFO 큐, hand, lazy promotion, quick demotion만으로 SIEVE를 구성한다. Evaluation: workloads=1,559 traces from 7 sources; 5 production cache libraries; baselines=ARC; 9 state-of-the-art algorithms; optimized LRU; metrics=miss ratio; throughput; integration LOC; results=Up to 63.2% lower miss than ARC; 2× LRU throughput; ≤20 LOC integration. Interpretation: 단순한 상태·지연 승격이 hit ratio와 멀티코어 확장성을 함께 준다. Reusable lesson: 생산 캐시 정책은 코드 복잡성과 lock 비용도 1급 지표로 보라. Applicability: 웹·객체 캐시 라이브러리. Limits: 작은 캐시의 scan형·일부 block workload에서는 LRU보다 나쁠 수 있고 ghost history가 없다.  +
sieve is simpler than lru an efficient turn key eviction algorithm for web caches 86f86293 +
Verification: abstract_only; confidence=high. Canonical title: SIEVE is Simpler than LRU: an Efficient Turn-Key Eviction Algorithm for Web Caches Question: 웹 캐시에 쉽게 이식하면서 LRU보다 단순하고 효율적인 축출 정책을 만들 수 있는가? Context: 최신 정책은 복잡하고 생산 캐시 라이브러리에 넣기 어렵다. Method: FIFO 큐, hand, lazy promotion, quick demotion만으로 SIEVE를 구성한다. Evaluation: workloads=1,559 traces from 7 sources; 5 production cache libraries; baselines=ARC; 9 state-of-the-art algorithms; optimized LRU; metrics=miss ratio; throughput; integration LOC; results=Up to 63.2% lower miss than ARC; 2× LRU throughput; ≤20 LOC integration. Interpretation: 단순한 상태·지연 승격이 hit ratio와 멀티코어 확장성을 함께 준다. Reusable lesson: 생산 캐시 정책은 코드 복잡성과 lock 비용도 1급 지표로 보라. Applicability: 웹·객체 캐시 라이브러리. Limits: 작은 캐시의 scan형·일부 block workload에서는 LRU보다 나쁠 수 있고 ghost history가 없다.  +
원본 주소 "https://s3wiki.yonsei.ac.kr/index.php?title=속성:Evidence_note&oldid=161"