본문으로 이동

속성:Observation

S3 연구 메모리

Text

직접 확인한 결과를 적습니다. 원인에 대한 해석은 따로 적습니다.

( | ) (20 | 50 | 100 | 250 | 500) 보기
이 속성을 사용하는 문서 20개를 보여줍니다.
s
scale and flexibility in distribution of hot content a83ac58f +
USENIX reports more than 800 PB distributed per day, 2–3× faster downloads than BitTorrent and Meta’s prior decentralized static tree, and production support for 106 use cases using 55 policies.  +
scalecache a scalable page cache for multiple solid state drives 60b5c227 +
workloads=file-system workloads on 64 cores and eight NVMe SSDs; baselines=Linux page cache; prior scalable page-cache scheme; metrics=file-system performance; scalability; writeback throughput; results=up to 6.81x over Linux; up to 4.50x over prior scalable scheme  +
scaleswap a scalable os swap system for all flash swap arrays 4041bd54 +
workloads=swap workloads on 128 cores and 8 NVMe SSDs; baselines=Linux, TMO, ExtMEM; metrics=throughput and average latency; results=up to 3.4x throughput and 11.5x lower latency vs Linux; +64% vs TMO; up to 5x ExtMEM  +
scalio scaling up dpu based jbof key value store with nvme of target offload 8db39764 +
workloads=high-density NVMe SSD configurations; baselines=conventional NVMe-oF target architecture; metrics=throughput and linearizability; results=up to 3.3x throughput  +
scheduler support for video oriented multimedia on client side virtualization c8366be4 +
frame-rate 추정오차는 VLC HighRes 0.79%, Quake3 3.05%, 간섭 중 VLC LowRes 0.55%로 보고됐다. CPU-intensive 경쟁 VM이 있어도 VLC HighRes가 최대 23.976 FPS에 가까운 평균과 낮은 분산을 유지했고, Quake3에도 적절한 share를 배정했다.  +
scoz a systemwide causal profiler for multicore systems 7b5e72e2 +
multithread workload에서 COZ와 동일 병목 및 유사한 잠재 speedup을 찾음. Dbench/Filebench의 ext4 kernel 병목은 기존 OS scalability 연구와 일치. NAS FT에서 transblock=128(작업집합 256 KB, 실험 CPU L2와 동일) 최적화로 throughput 3.48% 향상했고 실제 speedup 추세가 SCOZ 가상 speedup과 대부분 겹침.  +
selective memory deduplication for cost efficiency in mobile smart devices 9f3139a0 +
프로토타입은 기존 접근과 같은 메모리 절감을 제공하면서 계산비용을 크게 줄였고, 확보한 메모리로 앱 실행시간도 개선했다. 공식 초록에는 수치가 없다.  +
separate semantic replication hashes from provenance bound artifact hashes e7ab7a4e +
Whole artifact hashes differ when provenance differs. The versioned canonical projection without a trailing newline matches across a03 and a04: Qwen aa47da2041cdf33717348eba63a4c2f6f93e1b756888d3ed66a0d3739fa73894; Mistral 81fabcf6ac75f0f7566fbca184e0f46fcb67f70e036860a363e85f0175339cfa.  +
serverless in the wild characterizing and optimizing the serverless workload at a large cloud pr f2fa1121 +
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.  +
serverlessllm low latency serverless inference for large language models 69b8cd6d +
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  +
shape adaptive attention negative 20260722 +
The adaptive-policy surface had almost no exploitable headroom: the best deployable unfused policy improved over global by only 0-0.305%, the best fused policy improved by 0%, and the selected global candidate was already within 0.149-0.670% of the per-invocation oracle. Same-tile eager fusion looked large (2.280-3.333x), but a separate L40S CUDA Graph proxy reduced fused-versus-unfused replay speedup to 1.024-1.059x across four working-set regimes. The earlier analytical model also overcharged HBM traffic for L2-resident RoPE tensors and attributed 54-82% of chat-length savings to launch elimination, exactly the component CUDA Graphs amortize.  +
shiftlock mitigate one sided rdma lock contention via handover a5d67e63 +
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  +
sieve is simpler than lru an efficient turn key eviction algorithm for web caches 86f86293 +
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.  +
silk preventing latency spikes in log structured merge key value stores 8576916f +
workloads=Nutanix production and synthetic key-value workloads; baselines=RocksDB and TRIAD; metrics=p99 latency and throughput; results=up to 100x lower p99 with no significant penalty elsewhere  +
siloz leveraging dram isolation domains to prevent inter vm rowhammer 9d15762c +
workloads=cloud workloads; SPEC CPU2017; PARSEC 3; baselines=Linux/KVM; metrics=runtime overhead; isolation; results=Performance within ±0.5% of baseline.  +
smartlmk a memory reclamation scheme for improving user perceived app launch time d60522c6 +
실제 Android smartphone에서 user-perceived app launch time을 최대 13.2% 개선했다.  +
specinfer accelerating generative large language model serving with tree based speculative infer e609aa25 +
workloads=distributed LLM inference; offloaded LLM inference; baselines=existing LLM serving systems; metrics=inference speedup; accepted speculative tokens; output-distribution fidelity; results=1.5–2.8x distributed speedup; 2.6–3.5x offloading speedup  +
speculative multi level access in lsm tree based kv store 72e42f29 +
현실적 KV workload에서 기존 방식 대비 throughput 최대 85% 증가. 평균 latency 최대 45% 감소.  +
speed is all you need on device acceleration of large diffusion models via gpu aware optimizatio c53959fd +
workloads=Stable Diffusion 1.4, 512×512 image, 20 denoising steps; Samsung Galaxy S23 Ultra(Adreno 740)와 iPhone 14 Pro Max(A16); baselines=public repository implementation; metrics=UNet latency, intermediate tensor memory, initialized weight memory, end-to-end latency; results=S23 UNet latency 1098→525 ms, iPhone 1554→1043 ms; tensor memory 105→84 MB, weight memory 1640→2093 MB. S23의 20-step end-to-end 생성은 11.5초이며 INT8 quantization을 사용하지 않는다.  +
ssdstreamer specializing i o stack for large scale machine learning 6b597ff1 +
Apache Spark의 최선 기존 caching policy보다 성능을 유의하게 개선하면서 DRAM 용량은 1/4만 사용했다고 공식 초록이 보고한다. 공개 초록에는 정확한 속도 향상 수치가 없다.  +
원본 주소 "https://s3wiki.yonsei.ac.kr/index.php?title=속성:Observation&oldid=147"