속성:Observation
직접 확인한 결과를 적습니다. 원인에 대한 해석은 따로 적습니다.
m
확인된 사실은 Jinkyu Jeong·Hwanju Kim·Euiseong Seo·Joonwon Lee의 poster가 2012년 7월 서울에서 열린 ApSys 프로그램에 해당한다는 점이다. 메모리 절감량, CPU 비용, latency와 workload는 확인되지 않았다. +
기존 최신 partial mirroring 대비 시스템 FIT를 최대 19,000배 낮춤.
성능 오버헤드 3% 미만이며, 여러 경우 full mirroring에 근접한 신뢰성을 보고. +
midas minimizing write amplification in log structured systems through adaptive group number and 1cac80b9 +
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. +
minflow high performance and cost efficient data passing for i o intensive stateful serverless a 886ce161 +
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. +
workloads=16 reproduced real-world overload cases across MySQL, Apache, PostgreSQL, Elasticsearch, Solr, and etcd; baselines=non-overloaded execution, Protego, pBox, DARC, and PARTIES; metrics=normalized throughput, normalized p99 latency, request-drop rate, SLO attainment; results=Atropos sustains average normalized throughput 0.96 and average normalized p99 latency 1.16 while dropping fewer than 0.01% of requests. It meets the SLO in 14/16 cases; the reported multi-objective policy reduces normalized throughput by 10.2% relative to its performance-priority setting in the evaluated trade-off. +
mitigating resource usage dependency in sorting based kv stores on hybrid storage devices via op 31cbb8ad +
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 +
workloads=SPEC CPU; GAP graph workloads; Jailbreak and performance attacks; baselines=Panopticon; PRAC+ABO configurations; metrics=maximum aggressor activations; safe TRH; slowdown; SRAM overhead; results=Panopticon reaches 1,150 activations at threshold 128; MOAT ATH=64 safe for TRH=99; 0.28% average slowdown; 7 bytes SRAM per bank +
workloads=Mixtral 8x7B on one T4 16 GB; Mixtral 8x22B; DBRX; 2–4 low-cost GPUs; baselines=state-of-the-art offloading systems; FlexGen; metrics=throughput; CPU memory; resource utilization; results=up to 10.3x throughput; throughput bound with 2–3x less CPU memory +
MPI 기반 HPCG에서 예측치를 검증하고 특정 병목 지점을 실제로 최적화한 사례를 제시한다. 공개 초록에는 오차나 속도 향상 수치가 없다. +
n
workloads=LLaMA-2-70B, Mixtral-8x7B, LLaMA-3-8B; baselines=state-of-the-art LLM serving systems and analytical optimum; metrics=serving throughput and percent of optimum; results=up to 1.91x; 50–72% of optimum +
context 미사용 평균 Recall@1/2/3/4/5는 42.79/59.67/69.40/75.20/78.90%; context 사용 시 40.36/57.18/66.93/73.11/77.13%로 오히려 하락.
최고 월의 bidirectional model Recall@1은 50.79%, Recall@5는 86.55%; NAP 86.42%, AppUsage2Vec 85.93%, FALCON 80% 등과 비교.
긴 history보다 소수의 직전 앱이 더 유용한 경향을 보고. +
workloads=microbenchmarks; real applications under memory pressure; baselines=Linux TPP; hardware-assisted sampling approach; metrics=application performance; migration overhead; fault latency; results=up to 6x over Linux TPP under memory pressure +
not a dpu in name only unleashing rdma capable dpus in multi tenant serverless clouds with nadin 7031318c +
workloads=serverless ingress/data-plane workloads; baselines=host-centric serverless networking; metrics=requests/s, latency, host CPU cores; results=20.9x RPS; up to 21x lower latency; saves up to 7 CPU cores with 2 DPU cores +
확인된 사실은 우지원·이규선·정진규의 논문이 KCC 2019 학술발표논문집 1464–1466쪽에 수록됐다는 점이다. 장치, workload, baseline, fairness·latency·throughput 결과는 확인되지 않았다. +
공식 초록은 TCP/IP remote read와 compute-storage node 간 network traffic을 줄여 MinIO 성능을 개선했다고 보고한다.
공개 초록에는 개선율·latency·throughput 수치가 없다. +
확인된 사실은 김성환·이규선·정진규의 논문이 KCC 2019 학술발표논문집 1432–1434쪽에 수록됐다는 점이다. 장치, I/O 크기, CPU 사용량, latency·throughput 결과는 확인되지 않았다. +
o
workloads=I/O and data-processing workloads on near-storage accelerators; baselines=non-collaborative caching designs; metrics=I/O performance; data-processing performance; results=Up to 3.24× I/O and 3.06× processing performance. +
workloads=Apache AsterixDB LSM experiments; baselines=alternative merge schedulers and compaction policies; metrics=write stalls; throughput variance; I/O bandwidth; results=no single quantitative headline in primary abstract +
optimizing file systems on heterogeneous memory by integrating dram cache with virtual memory ma aaa7aa17 +
workloads=microbenchmarks; real applications; baselines=DAX; cache-based file systems; metrics=application performance; results=Up to two orders of magnitude in microbenchmarks; 10.6× vs DAX and 9.9× vs cache-based FS. +
총 startup time을 35% 줄였다. +