속성으로 검색

이 문서는 속성과 이름이 지정된 값으로 설명된 개체를 찾기 위한 간단한 탐색 인터페이스를 제공합니다. 그 밖에 사용 가능한 검색 인터페이스에는 문서 속성 검색 및 Ask 쿼리 빌더가 있습니다.

결과 목록

• Lesson:dwkv collinear slo confound  + (SLO scheduling, ablation design, causal policy comparison, simulator workload generation)
• Lesson:technical review demand based coordinated scheduling for smp vms 3d4e78ab  + (SMP VM의 parallel runtime, multithreaded service, lock/communication-heavy workload 스케줄링에 해당한다.)
• Lesson:demand based coordinated scheduling for smp vms a57bd76a  + (SMP VM의 parallel runtime, multithreaded service, lock/communication-heavy workload 스케줄링에 해당한다.)
• Lesson:efficient memory overcommitment for i o passthrough enabled vms via fine grained page meta data d32f4845  + (SR-IOV·passthrough VM 클라우드. Limits: 게스트·하이퍼바이저·장치 간 협조와 메타데이터 유지가 필요하다.)
• Lesson:rl watchdog a fast and predictable ssd liveness watchdog on storage systems a9a43b61  + (SSD and RAID liveness monitoring. Limits: Generalization depends on training, SSD models, RAID layouts, and fault types represented in evaluation.)
• Lesson:ionia high performance replication for modern disk based kv stores 7cafc97e  + (SSD 기반 복제 KV 저장소. Limits: WO-KV 의미론과 bounded history에 의존하며 일부 읽기는 추가 왕복으로 fallback한다.)
• Lesson:mitigating resource usage dependency in sorting based kv stores on hybrid storage devices via op 31cbb8ad  + (SSD-backed ordered key-value stores with LSM-style indexing. Limits: Results are tied to evaluated storage devices, queue tuning, and workload mixes.)
• Lesson:achieving low latency graph based vector search via aligning best first search algorithm with ss 77c9550f  + (SSD-resident graph ANN indexes at billion-vector scale. Limits: Reported relation to in-memory Vamana is still slower; gains depend on graph/search and SSD characteristics.)
• Lesson:polymorphic error correction 25fcf9e1  + (Secure DDR5 memory systems combining ECC and integrity checks. Limits: Assumes a MAC+ECC secure-memory setting; performance, iteration latency, and all fault assumptions were not verified from full text.)
• Lesson:faasmem improving memory efficiency of serverless computing with memory pool architecture b6ab9eec  + (Serverless/FaaS clusters with a remote memory pool. Limits: Relies on function lifetime/access regularity, remote-memory bandwidth, and evaluated serverless workloads.)
• Lesson:bypassd enabling fast userspace access to shared ssds da9e5d79  + (Shared NVMe SSDs serving unmodified storage applications. Limits: Depends on suitable IOMMU/device support and a setup/control path in the OS.)
• Lesson:loom efficient capture and querying of high frequency telemetry bbcbd611  + (Single-host recent high-frequency telemetr … Single-host recent high-frequency telemetry and interactive debugging.</br></br>Limits: Loom targets recent situational analysis rather than long-term storage, has finite ingest capacity, and may lose the active in-memory block (for example 64 MiB) on machine failure.k (for example 64 MiB) on machine failure.)
• Lesson:twinpilots a new computing paradigm for gpu cpu parallel llm inference aa39d085  + (Single-server LLM inference with GPU-memory-constrained models. Limits: Evidence is tied to one CPU/GPU server and selected models; gains depend on operator balance and host memory bandwidth.)
• Lesson:speed is all you need on device acceleration of large diffusion models via gpu aware optimizatio c53959fd  + (Stable Diffusion 1.4 계열 512×512 on-device inference와 유사한 모바일 GPU kernel 최적화에 적용합니다. Limits: S23 Ultra와 iPhone 14 Pro Max, 특정 모델·해상도·20-step 설정 중심이며 에너지·품질·다른 SoC 일반화는 추가 평가가 필요합니다.)
• Lesson:xrp in kernel storage functions with ebpf 49cdd043  + (Storage engines with short, data-dependent lookup chains on NVMe. Limits: Logic is constrained by eBPF verification and kernel-state interfaces; exact evaluation details were not extracted.)
• Lesson:phasedrr read reclaim scheduling without page level access counting 69578f95  + (TLC/QLC SSD FTL read-reclaim. Limits: SSDsim 기반이며 세 window·threshold와 trace hotness에 민감하다.)
• Lesson:inf2 high throughput generative inference of large language models using near storage processing 3fcead73  + (Throughput 중심의 offline inference, 수만~128K … Throughput 중심의 offline inference, 수만~128K token의 장문맥, 가중치와 KV cache가 GPU 또는 host DRAM에 여유 있게 들어가지 않는 decoder-only MHA/GQA/MoE 모델, batch benchmark와 대규모 정보 추출에 적용할 수 있습니다. Commercial SmartSSD 또는 programmable near-storage accelerator가 필요합니다. CXL 기반 near-data system에도 host traffic 최소화, host/device cooperative work, exact streaming attention이라는 원칙은 재사용할 수 있지만 HILOS 구현 자체는 PCIe 기반입니다.</br></br>Limits: strict-latency online serving의 TTFT/TPOT를 검증하지 않았습니다. 큰 성능 향상은 4–16개 SmartSSD 병렬 구성에 의존하며 전용 expansion chassis, GPUDirect Storage와 FPGA software stack이 필요합니다. 일부 모델의 128K 실험은 pretraining context를 넘으므로 system scaling만 검증하고 답변 품질을 보장하지 않습니다. 기본 FP16, batch 16, output 64 결과는 workload에 따라 달라집니다. DRAM이 충분하면 FLEX(DRAM)이 더 비용 효율적입니다. DeepSpeed 비교는 저자 추가 UVM 확장이고, conventional SSD energy는 측정값이 아니라 datasheet 값입니다. PCIe 5.0 SSD를 따라가려면 2,000개가 넘는 DSP가 필요하다는 분석처럼 현 FPGA에는 확장 한계가 있습니다. Softmax가 큰 attention group에서 실행 시간의 50% 이상을 차지하며, capacity 대비 bandwidth 불균형으로 SmartSSD당 4TB 중 평가 peak 사용량은 600GB 미만입니다. Cost와 endurance는 장기 운영 관측이 아니라 논문의 가격·PBW·retention 가정에 따른 모델입니다.관측이 아니라 논문의 가격·PBW·retention 가정에 따른 모델입니다.)
• Lesson:telescope telemetry for gargantuan memory footprint applications 6a766ffb  + (Tiering and telemetry for very large-memory applications. Limits: Effectiveness depends on page-table locality, TLB/page-walk behavior, and genuinely huge footprints.)
• Lesson:gqa training generalized multi query transformer models from multi head checkpoints a6839ee2  + (Transformer 생성 추론과 KV 캐시 절감. Limits: 그룹 수 선택과 5% 추가 학습비가 필요하며 모델군별 일반화가 제한될 수 있다.)
• Lesson:utopia fast and efficient address translation via hybrid restrictive flexible virtual to physica 796873cf  + (Translation-intensive CPUs and large-memory applications. Limits: Results rely on architectural simulation and restrictive-segment allocation; sharing and fragmentation can reduce eligibility.)
• Lesson:technical review ufs a8ea798b  + (UFS를 탑재한 메모리 제약형 모바일·엣지 장치의 LLM 추론.)
• Lesson:ufs bbe26b78  + (UFS를 탑재한 메모리 제약형 모바일·엣지 장치의 LLM 추론.)
• Lesson:rfuse modernizing userspace filesystem framework through scalable kernel userspace communication b3196db3  + (Userspace filesystems whose daemon logic can remain unchanged. Limits: Exact workloads, hardware, and quantitative comparisons were not extracted from full text.)
• Lesson:technical review virtual asymmetric multiprocessor for interactive performance of consolidated d 70c3d6f5  + (VDI·멀티테넌트 데스크톱에서 대화형 QoS와 백그라운드 처리량의 명시적 trade-off를 설계할 때 직접 적용 가능하다.)
• Lesson:virtual asymmetric multiprocessor for interactive performance of consolidated desktops ec394686  + (VDI·멀티테넌트 데스크톱에서 대화형 QoS와 백그라운드 처리량의 명시적 trade-off를 설계할 때 직접 적용 가능하다.)
• Lesson:managing memory tiers with cxl in virtualized environments 357c7cc7  + (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.)

• Lesson:d2fq device direct fair queueing for nvme ssds 36293af6  + (WRR을 실제 구현한 NVMe SSD를 공유하는 multi-tenant Linux I/O.)
• Lesson:technical review d2fq device direct fair queueing for nvme ssds c60d73e1  + (WRR을 실제 구현한 NVMe SSD를 공유하는 multi-tenant Linux I/O.)

• Lesson:technical review task aware virtual machine scheduling for i o performance 26a8da47  + (Xen형 hypervisor에서 mixed CPU/I/O workloads의 tail/response latency 개선에 해당한다.)
• Lesson:task aware virtual machine scheduling for i o performance f431ccbe  + (Xen형 hypervisor에서 mixed CPU/I/O workloads의 tail/response latency 개선에 해당한다.)
• Lesson:research autopilot 20260723t090001z-gpu  + (algebraic-ml-compiler: Only the recorded shapes, dtypes, software revision, and physical L40S GPU 3.)
• Lesson:research autopilot 20260720t090001z  + (alpha-factory: Holds for gated candidate-s … alpha-factory: Holds for gated candidate-selection pipelines that filter per-candidate on standalone merit before assembling a portfolio or ensemble, where some candidate roles are expected to underperform standalone benchmarks. Demonstrated on one ETF macro pool, one asset universe, and a single 587-observation validation window with lambda=0.5 tested rather than calibrated; the dose-response shape, generalization beyond two lineages, and out-of-sample confirmation on the sealed hidden test are all unestablished. The joint-necessity structure is expected to transfer more readily than the specific effect sizes.</br>complex-nn-signal: Synthetic single-tone Doppler-bin classification, K=4 classes, sequence length L=16, complex AWGN at −5/0 dB, 5 seeds, CPU, small models with shared hyperparameters; detectors compared are a uniform-grid correlator scan and a diagonal unit-modulus linear complex recurrence with a linear magnitude read-out. The negative result is bounded by the learned bank having no cell structure, so it cannot implement a per-cell max even in principle — the fair retest is a max-pooled learned bank, ideally under a non-uniform within-cell prior where the uniform grid is provably mismatched. Conclusions about budget-matched evaluation and the train-accuracy underfitting diagnostic generalize beyond this task; the specific accuracy margins do not.</br>azure_inference_queueing: Holds for single-resource (memory-constrained) admission control where each item has one reduced-quality variant with a proportional weight multiplier m and a fractional value loss a satisfying m + a < 1, compared against the fractional LP relaxation rather than a full-horizon MDP optimum. The inertness result (m <= 1/2) is a general property of the increment weights and is workload-independent; the quantitative decoupling penalties are specific to workloads where accuracy is priced linearly and cheaply relative to revenue (validated on synthetic Azure NDm A100 v4 parameters and real BurstGPT windows, 2-3 service classes). Untested for three or more variants, convex or SLA-cliff accuracy penalties, and — most importantly — multi-resource settings: the model captures only quantization's memory effect, not its compute effect (INT8 = 2x TFLOPS), which could plausibly change the decoupling conclusion under a compute-binding regime.</br>gaussian-3mul-compiler: IEEE-754 binary64 complex multiplication on CPU, measured against exact rational truth over uniform/normal/lognormal/mixed-scale/large-magnitude input distributions. The contraction result assumes the compiler fuses one product per output part (the asymmetric contraction numpy emits); the mirrored fusion choice has a distinct distribution-dependent profile and is not covered by the conclusion. Imaginary part only, since 3-mul and 4-mul real parts are bit-identical. The median-vs-mean estimator lesson generalises to any heavy-tailed floating-point error comparison; the specific spread magnitudes are sample-size dependent.pread magnitudes are sample-size dependent.)
• Lesson:research autopilot 20260721t210001z  + (alpha-factory: Offline portfolio-construct … alpha-factory: Offline portfolio-construction / ensemble-selection research where a treatment changes ensemble composition; single ETF-macro universe and one frozen candidate pool re-sliced by evaluation window (not re-evolved per fold), rolling-origin (not walk-forward-refit). Second-universe transfer (etf_macro_transfer) and hidden-test confirmation remain open. and hidden-test confirmation remain open.)
• Lesson:research autopilot 20260721t090001z  + (alpha-factory: Offline portfolio/backtest … alpha-factory: Offline portfolio/backtest research where a train/select/validate/hidden split is percentage-based and non-overlapping, and a headline OOS metric is computed on a window that does not touch the sealed suffix. The CSCV PBO gate is only weakly informative when the search collapses to few distinct strategies; use it as a sanity gate, not a full search-space overfitting test. not a full search-space overfitting test.)
• Lesson:research autopilot 20260723t030001z  + (alpha-factory: alpha-factory and similar e … alpha-factory: alpha-factory and similar evolutionary alpha-discovery pipelines with an eligibility/diversifier admission track gated on standalone fitness and correlation-to-elite. Bounded here to two universes from one Yahoo-adjusted ETF vendor snapshot and a seed-only pool (max_generations=0); an evolved pool or cross-vendor data could still admit a decorrelated fitness-positive crisis sleeve.correlated fitness-positive crisis sleeve.)
• Lesson:research autopilot 20260721t150001z  + (azure_inference_queueing: Single-period LP … azure_inference_queueing: Single-period LP-vs-IP relaxation gap of a multidimensional (m-constraint) 0/1 knapsack with nonnegative values and weights; verified for m=2 (memory×compute) on CPU-synthetic instances. Does not extend to full-horizon/rolling optimality, arrival dynamics, real serving traces, or any Azure production/SLA claim, and does not certify any greedy policy.m, and does not certify any greedy policy.)
• Lesson:research autopilot 20260722t210001z  + (complex-nn-signal: Synthetic single-tone L … complex-nn-signal: Synthetic single-tone LFM chirp classification, CPU, K=4/L=16, one optimizer, descriptive SESOI verdicts, not externally preregistered and no captured data. The full escape holds at SNR=0; the SNR=-5 case only partially closes. The adaptive-gate lesson generalizes to models with an optional/nuisance DOF and a data-scaled loss gradient, but the specific accuracy numbers and the exact nulling lambda are task- and SNR-specific.nulling lambda are task- and SNR-specific.)
• Lesson:technical review rigorous rental memory management for embedded systems 1b4e69c5  + (contiguous device memory를 쓰는 embedded/mobile 시스템의 idle capacity 재활용에 해당한다.)
• Lesson:rigorous rental memory management for embedded systems 5632faf0  + (contiguous device memory를 쓰는 embedded/mobile 시스템의 idle capacity 재활용에 해당한다.)
• Lesson:dwkv doomed chase domino  + (deadline scheduling, overload control, best-effort admission, heuristic priority queues)
• Lesson:technical review appwatch detecting kernel bug for protecting consumer electronics applications 779775f3  + (device-driver 신뢰성이 낮은 embedded product의 critical-process protection에 해당한다.)
• Lesson:appwatch detecting kernel bug for protecting consumer electronics applications a27fb99c  + (device-driver 신뢰성이 낮은 embedded product의 critical-process protection에 해당한다.)
• Lesson:technical review transparent and selective real time interrupt services for performance improvem b78f264b  + (embedded Linux의 real-time interrupt handling과 throughput 보호에 해당한다.)
• Lesson:transparent and selective real time interrupt services for performance improvement 6b173779  + (embedded Linux의 real-time interrupt handling과 throughput 보호에 해당한다.)
• Lesson:technical review daac device reserved memory as an eviction based file cache f0f149f0  + (embedded/mobile reserved-memory reclamation과 secondary file cache에 직접 해당한다.)
• Lesson:daac device reserved memory as an eviction based file cache db4ed198  + (embedded/mobile reserved-memory reclamation과 secondary file cache에 직접 해당한다.)
• Lesson:dwkv failed run fail closed  + (experiment runners, benchmark data quality, failure recovery, reproducibility audits)
• Lesson:halfmoon log optimal fault tolerant stateful serverless computing e44c470a  + (fault-tolerant stateful FaaS. Limits: 외부 상태 의미론과 workload 분류, 프로토콜 전환 정확성에 의존한다.)
• Lesson:the design and implementation of a capacity variant storage system 91c1817e  + (flash SSD와 로그 구조 파일시스템. Limits: 사용자 용량을 희생하며 custom SSD·FS·manager와 수명 가정이 필요하다.)
• Lesson:technical review energy efficient scheduling of real time tasks on multicore processors cc0782cd  + (homogeneous multicore real-time scheduler의 energy-aware partitioning/core parking에 해당한다.)
• Lesson:energy efficient scheduling of real time tasks on multicore processors a70b387a  + (homogeneous multicore real-time scheduler의 energy-aware partitioning/core parking에 해당한다.)
• Lesson:technical review z journal scalable per core journaling 1f19a370  + (many-core ext4-like journaling filesystem과 낮거나 중간 수준 공유의 metadata/fsync workload에 높음.)