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
sieve is simpler than lru an efficient turn key eviction algorithm for web caches 86f86293 +
Bibliographic paper record.  +
silk preventing latency spikes in log structured merge key value stores 8576916f +
Verification: official USENIX page and abstract; confidence=high. Canonical title: SILK: Preventing Latency Spikes in Log-Structured Merge Key-Value Stores Question: How can an LSM store prevent compaction from causing severe foreground tail-latency spikes? Context: Background compaction competes with user I/O and can create long stalls despite high average throughput. Method: SILK schedules internal I/O opportunistically, prioritizes lower LSM levels, and makes compaction preemptible in a RocksDB-derived store. Evaluation: 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 Interpretation: Tail latency depends on explicitly scheduling maintenance I/O, not merely tuning compaction thresholds. Reusable lesson: Make background storage work preemptible and prioritize tasks that avert future write stalls. Applicability: LSM key-value stores with latency-sensitive foreground traffic. Limits: Results target the evaluated RocksDB-derived implementation and storage/workload mixes.  +
silk preventing latency spikes in log structured merge key value stores 8576916f +
Bibliographic paper record.  +
silk preventing latency spikes in log structured merge key value stores 8576916f +
Verification: official USENIX page and abstract; confidence=medium. Canonical title: SILK: Preventing Latency Spikes in Log-Structured Merge Key-Value Stores Question: How can an LSM store prevent compaction from causing severe foreground tail-latency spikes? Context: Background compaction competes with user I/O and can create long stalls despite high average throughput. Method: SILK schedules internal I/O opportunistically, prioritizes lower LSM levels, and makes compaction preemptible in a RocksDB-derived store. Evaluation: 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 Interpretation: Tail latency depends on explicitly scheduling maintenance I/O, not merely tuning compaction thresholds. Reusable lesson: Make background storage work preemptible and prioritize tasks that avert future write stalls. Applicability: LSM key-value stores with latency-sensitive foreground traffic. Limits: Results target the evaluated RocksDB-derived implementation and storage/workload mixes.  +
siloz leveraging dram isolation domains to prevent inter vm rowhammer 9d15762c +
Bibliographic paper record.  +
siloz leveraging dram isolation domains to prevent inter vm rowhammer 9d15762c +
Verification: abstract_only; confidence=high. Canonical title: Siloz: Leveraging DRAM Isolation Domains to Prevent Inter-VM Rowhammer Question: DRAM 성능을 거의 희생하지 않고 VM 간 Rowhammer를 막을 수 있는가? Context: Rowhammer 완화는 큰 성능·용량 비용이 들거나 하드웨어 지원이 필요하다. Method: DRAM subarray group을 VM별 isolation domain으로 배정하고 EPT 무결성을 보호한다. Evaluation: workloads=cloud workloads; SPEC CPU2017; PARSEC 3; baselines=Linux/KVM; metrics=runtime overhead; isolation; results=Performance within ±0.5% of baseline. Interpretation: DRAM 내부 장애 범위를 VM 배치 단위로 쓰면 공격 전파를 저비용으로 차단할 수 있다. Reusable lesson: 보안 격리를 물리 메모리 고장 도메인과 정렬하라. Applicability: 멀티테넌트 x86 클라우드. Limits: 선정 Intel 세대와 추정 subarray/address mapping·BIOS 정보에 의존한다.  +
siloz leveraging dram isolation domains to prevent inter vm rowhammer 9d15762c +
Verification: abstract_only; confidence=medium. Canonical title: Siloz: Leveraging DRAM Isolation Domains to Prevent Inter-VM Rowhammer Question: DRAM 성능을 거의 희생하지 않고 VM 간 Rowhammer를 막을 수 있는가? Context: Rowhammer 완화는 큰 성능·용량 비용이 들거나 하드웨어 지원이 필요하다. Method: DRAM subarray group을 VM별 isolation domain으로 배정하고 EPT 무결성을 보호한다. Evaluation: workloads=cloud workloads; SPEC CPU2017; PARSEC 3; baselines=Linux/KVM; metrics=runtime overhead; isolation; results=Performance within ±0.5% of baseline. Interpretation: DRAM 내부 장애 범위를 VM 배치 단위로 쓰면 공격 전파를 저비용으로 차단할 수 있다. Reusable lesson: 보안 격리를 물리 메모리 고장 도메인과 정렬하라. Applicability: 멀티테넌트 x86 클라우드. Limits: 선정 Intel 세대와 추정 subarray/address mapping·BIOS 정보에 의존한다.  +
smartlmk a memory reclamation scheme for improving user perceived app launch time d60522c6 +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1.  +
smartlmk a memory reclamation scheme for improving user perceived app launch time d60522c6 +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_smartlmk_a_memory_reclamation_scheme_for_improving_user_perceived_app_launch_ti_e990e433. 확인 범위: official_abstract. 확인한 자료: https://doi.org/10.1145/2894755 ; https://yonsei.elsevierpure.com/en/publications/smartlmk-a-memory-reclamation-scheme-for-improving-user-perceived/. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다.  +
specinfer accelerating generative large language model serving with tree based speculative infer e609aa25 +
Verification: full_text; confidence=high. Question: Can speculative decoding exploit multiple draft continuations while exactly preserving the target LLM's output distribution? Context: A single small draft model can propose poor tokens, limiting acceptance and leaving target-model parallelism unused. Method: SpecInfer has multiple small models construct a token tree, then verifies tree nodes in parallel with the target LLM using distribution-preserving acceptance. Evaluation: 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 Interpretation: Branching speculation converts uncertain draft quality into parallel target-model verification. Reusable lesson: When cheap predictors are uncertain, aggregate diverse candidates and verify them in one expensive parallel pass. Applicability: Autoregressive LLM serving with distributed or offloaded target models. Limits: Benefit depends on draft accuracy, token-tree construction cost, target parallelism, model pair, and hardware.  +
specinfer accelerating generative large language model serving with tree based speculative infer e609aa25 +
Bibliographic paper record.  +
specinfer accelerating generative large language model serving with tree based speculative infer e609aa25 +
Verification: full_text; confidence=high. Question: Can speculative decoding exploit multiple draft continuations while exactly preserving the target LLM's output distribution? Context: A single small draft model can propose poor tokens, limiting acceptance and leaving target-model parallelism unused. Method: SpecInfer has multiple small models construct a token tree, then verifies tree nodes in parallel with the target LLM using distribution-preserving acceptance. Evaluation: 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 Interpretation: Branching speculation converts uncertain draft quality into parallel target-model verification. Reusable lesson: When cheap predictors are uncertain, aggregate diverse candidates and verify them in one expensive parallel pass. Applicability: Autoregressive LLM serving with distributed or offloaded target models. Limits: Benefit depends on draft accuracy, token-tree construction cost, target parallelism, model pair, and hardware.  +
speculative multi level access in lsm tree based kv store 72e42f29 +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1.  +
speculative multi level access in lsm tree based kv store 72e42f29 +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_speculative_multi_level_access_in_lsm_tree_based_kv_store_733f947b. 확인 범위: official_abstract. 확인한 자료: https://yonsei.elsevierpure.com/en/publications/speculative-multi-level-access-in-lsm-tree-based-kv-store/ ; https://doi.org/10.1109/LCA.2022.3219808. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다.  +
speed is all you need on device acceleration of large diffusion models via gpu aware optimizatio c53959fd +
Verification: abstract_only; confidence=high. Canonical title: Speed Is All You Need: On-Device Acceleration of Large Diffusion Models via GPU-Aware Optimizations Question: 대형 확산 모델을 양자화 없이 스마트폰 GPU에서 실용 속도로 실행할 수 있는가? Context: Stable Diffusion의 반복 UNet 실행과 GPU 비효율이 모바일 지연을 키운다. Method: GPU 커널·메모리·그래프 실행을 장치 특성에 맞게 최적화한다. Evaluation: workloads=Stable Diffusion 1.4, 512×512, 20 steps; Samsung Galaxy S23 Ultra; baselines=; metrics=end-to-end latency; results=Under 12 seconds without INT8 quantization. Interpretation: 모델 변경 없이 구현 계층 최적화만으로도 모바일 생성 지연을 크게 줄일 수 있다. Reusable lesson: 모바일 ML은 연산량뿐 아니라 GPU 실행·메모리 경로를 공동 최적화하라. Applicability: 온디바이스 이미지 생성. Limits: 단일 모델·주력 장치 중심이고 품질·에너지의 폭넓은 비교가 제한적이다.  +
speed is all you need on device acceleration of large diffusion models via gpu aware optimizatio c53959fd +
Bibliographic paper record.  +
speed is all you need on device acceleration of large diffusion models via gpu aware optimizatio c53959fd +
Verification: abstract_only; confidence=medium. Canonical title: Speed Is All You Need: On-Device Acceleration of Large Diffusion Models via GPU-Aware Optimizations Question: 대형 확산 모델을 양자화 없이 스마트폰 GPU에서 실용 속도로 실행할 수 있는가? Context: Stable Diffusion의 반복 UNet 실행과 GPU 비효율이 모바일 지연을 키운다. Method: GPU 커널·메모리·그래프 실행을 장치 특성에 맞게 최적화한다. Evaluation: workloads=Stable Diffusion 1.4, 512×512, 20 steps; Samsung Galaxy S23 Ultra; baselines=; metrics=end-to-end latency; results=Under 12 seconds without INT8 quantization. Interpretation: 모델 변경 없이 구현 계층 최적화만으로도 모바일 생성 지연을 크게 줄일 수 있다. Reusable lesson: 모바일 ML은 연산량뿐 아니라 GPU 실행·메모리 경로를 공동 최적화하라. Applicability: 온디바이스 이미지 생성. Limits: 단일 모델·주력 장치 중심이고 품질·에너지의 폭넓은 비교가 제한적이다.  +
ssdstreamer specializing i o stack for large scale machine learning 6b597ff1 +
Source: Yonsei University Computer Systems Laboratory publication list supplied by the user. Manifestation 1 of 1.  +
ssdstreamer specializing i o stack for large scale machine learning 6b597ff1 +
정본 Lesson 보강 근거. 검토 원본: Lesson:technical_review_ssdstreamer_specializing_i_o_stack_for_large_scale_machine_learning_5d1f4a63. 확인 범위: official_abstract. 확인한 자료: https://doi.org/10.1109/MM.2019.2930497 ; https://yonsei.elsevierpure.com/en/publications/ssdstreamer-specializing-io-stack-for-large-scale-machine-learnin/. 질문, 방법, 평가, 해석, 재사용 교훈, 적용 범위와 한계를 같은 Lesson 본문에 통합했습니다.  +
stem streaming based fpga acceleration for large scale compactions in lsm kv d204166c +
Verification: official_abstract; confidence=medium. Question: How can FPGA compaction scale to large LSM runs without materializing oversized intermediate data? Context: Large-scale merge/compaction is CPU- and I/O-intensive, and rigid FPGA pipelines scale poorly across run sizes. Method: STEM uses clean-cut streaming compaction, a multi-unit FPGA pipeline, and dynamic pipeline scheduling. Evaluation: workloads=large-scale LSM compactions; baselines=RocksDB compaction; state-of-the-art FPGA compaction accelerator; metrics=compaction speedup; pipeline utilization; results=27x average and 35x maximum vs RocksDB; 2.09–2.27x vs prior FPGA accelerator Interpretation: Streaming boundaries and elastic pipeline allocation are central to sustaining accelerator utilization. Reusable lesson: Design data-reorganization accelerators around streaming partitions and dynamically scheduled units. Applicability: Large-scale LSM compaction on FPGA-enabled storage servers. Limits: Benefits depend on FPGA resources, host/storage balance, and the evaluated compaction distributions.  +
원본 주소 "https://s3wiki.yonsei.ac.kr/index.php?title=속성:Evidence_note&oldid=161"