속성으로 검색
외관
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:technical review a fully associative tagless dram cache 8377cc32 + (Cache granularity를 OS page와 맞추고 cTLB가 virtual-to-physical 대신 virtual-to-cache mapping을 저장한다. TLB miss handler가 page table과 cTLB를 갱신하며 cTLB에서 밀린 페이지는 남은 공간의 victim cache로 사용한다.)
- Lesson:a fully associative tagless dram cache f6383680 + (Cache granularity를 OS page와 맞추고 cTLB가 virtual-to-physical 대신 virtual-to-cache mapping을 저장한다. TLB miss handler가 page table과 cTLB를 갱신하며 cTLB에서 밀린 페이지는 남은 공간의 victim cache로 사용한다.)
- Lesson:cacheblend fast large language model serving for rag with cached knowledge fusion 056a5f5e + (CacheBlend selectively recomputes fewer than 15% of tokens in typical cases and pipelines recomputation with cache fetch.)
- Lesson:cost efficient large language model serving for multi turn conversations with cachedattention aa200b5c + (CachedAttention stores KV state hierarchically, preloads and saves it layer by layer, coordinates fetch/eviction with scheduling, and decouples position encoding from truncation.)
- Lesson:citadel rethinking memory allocation to safeguard against inter domain rowhammer exploits 8fd39aca + (Citadel allocates flexible security domains into physically disjoint DRAM regions through a redesigned allocator.)
- Lesson:tiered memory management access latency is the key aaf24437 + (Colloid estimates each tier's loaded latency from hardware counters and Little's Law, then moves pages to balance average access latency; it integrates with HeMem, TPP, and MEMTIS.)
- Lesson:achieving microsecond scale tail latency efficiently with approximate optimal scheduling e8d8e302 + (Concord는 응용 비종속 근사 최적 스케줄링으로 다중 큐·비엄격 선점을 사용한다.)
- Lesson:how to copy memory coordinated asynchronous copy as a first class os service 8e43b859 + (Copier coordinates copies asynchronously, selects hardware such as SIMD/DMA, and absorbs redundant intermediate copies.)
- Lesson:cornus atomic commit for a cloud dbms with storage disaggregation 279df452 + (Cornus는 저장소의 원자적 LogOnce(CAS) 연산으로 참가자 로그를 조정해 코디네이터 결정 로그를 없앤다.)
- Lesson:corobase coroutine oriented main memory database engine c4a28018 + (CoroBase expresses each transaction as a coroutine and interleaves transactions to batch memory accesses and issue software prefetches while retaining the transaction API.)
- Lesson:crossprefetch accelerating i o prefetching for modern storage 59296dd1 + (CrossPrefetch exports runtime state to the OS, separates demand/prefetch paths, tracks fine-grained indices for shared files, and adapts prediction aggressiveness.)
- Lesson:advocating for key value stores with workload pattern aware dynamic compaction 9489efc9 + (DOPA-DB는 RocksDB 위에 여러 key range를 둔 tiered LSM을 구성하고, L0 크기·flush 속도·누적 compaction 속도를 모니터링하는 Compaction Size Recommender로 upper-level compaction 크기를 workload phase에 맞춰 바꾼다.)
- Lesson:siloz leveraging dram isolation domains to prevent inter vm rowhammer 9d15762c + (DRAM subarray group을 VM별 isolation domain으로 배정하고 EPT 무결성을 보호한다.)
- Lesson:technical review ssdstreamer specializing i o stack for large scale machine learning 5d1f4a63 + (DRAM을 상위 캐시로 쓰지 않고 coarse-grained prefetch용 stream buffer로 사용하며, 큰 SSD cache는 무거운 커널 I/O 경로를 우회하는 전용 user-space I/O stack 위에서 구동한다.)
- Lesson:ssdstreamer specializing i o stack for large scale machine learning 6b597ff1 + (DRAM을 상위 캐시로 쓰지 않고 coarse-grained prefetch용 stream buffer로 사용하며, 큰 SSD cache는 무거운 커널 I/O 경로를 우회하는 전용 user-space I/O stack 위에서 구동한다.)
- Lesson:mitigating resource usage dependency in sorting based kv stores on hybrid storage devices via op 31cbb8ad + (DecouKV uses mergeable skip-list indexes, append-only value data, elastic level capacities, and parameterized request queues.)
- Lesson:deltazip efficient serving of multiple full model tuned llms 8420449f + (DeltaZip stores a shared base plus structured-sparse, quantized, optionally lossless deltas and executes custom sparse-batched matrix multiplication in a vLLM-based server.)
- Lesson:technical review efficient hybrid polling for ultra low latency storage devices 545391d6 + (EHP는 device service time과 queueing delay를 분리 추정해 sleep 시간을 적응시키고, 긴 I/O에는 polling이 이득이 없으면 interrupt로 전환한다. Linux 4.16.1에서 Z-SSD와 Optane으로 평가했다.)
- Lesson:efficient hybrid polling for ultra low latency storage devices e753019a + (EHP는 device service time과 queueing delay를 분리 추정해 sleep 시간을 적응시키고, 긴 I/O에는 polling이 이득이 없으면 interrupt로 전환한다. Linux 4.16.1에서 Z-SSD와 Optane으로 평가했다.)
- Lesson:extending applications safely and efficiently 7a5f9f65 + (EIM makes extension capabilities/resources explicit; bpftime enforces them with eBPF verification, memory protection keys, and dynamic rewriting.)
- Lesson:emt an os framework for new memory translation architectures 6d3356f6 + (EMT provides an architecture-neutral Linux framework for radix/hash translation and hardware optimizations, and ports Linux memory management to support ECPT and FPT.)
- Lesson:technical review nvme driven lazy cache coherence for immutable data with nvme over fabrics 4d6bc037 + (EXT4-oF는 EXT4를 NVMe-oF 공유 장치로 확장하고, 파일 생성 후 다른 node가 false-negative open을 만나면 NVMe-oF에서 metadata를 다시 읽는 NVMe-driven lazy coherence를 사용한다.)
- Lesson:nvme driven lazy cache coherence for immutable data with nvme over fabrics c19688e2 + (EXT4-oF는 EXT4를 NVMe-oF 공유 장치로 확장하고, 파일 생성 후 다른 node가 false-negative open을 만나면 NVMe-oF에서 metadata를 다시 읽는 NVMe-driven lazy coherence를 사용한다.)
- Lesson:exploring the asynchrony of slow memory filesystem with easyio 0952deee + (EasyIO uses an on-chip DMA engine for asynchronous data movement, user-level task scheduling, orderless parallel data/metadata updates, two-stage locking with hardware-assisted unlock, and adaptive DMA-channel/QoS control.)
- Lesson:how storage i o affects user perceived latency in mobile apps b59ccb75 + (EuroSys 2015 공식 poster 목록을 확인했지만 poster PDF, 공식 초록, DOI와 artifact를 찾지 못했다. 측정 장치, 앱 선정, I/O 추적 방식과 지연 분해 방법은 미확인이다.)
- Lesson:evendb optimizing key value storage for spatial locality 7beea059 + (EvenDB는 공간 파티셔닝과 LSM batch I/O를 결합한다.)
- Lesson:extmem enabling application aware virtual memory management for data intensive applications 8a6dc0e4 + (ExtMem provides a Linux-compatible, backend-agnostic framework for application policies, multithreaded execution, telemetry, and incremental deployment.)
- Lesson:technical review efficient footprint caching for tagless dram caches 5090e3ab + (F-TDC는 cached page footprint를 cache-tag array가 아니라 TLB에서 추적하고 eviction 시 footprint history를 제한된 on-chip table 대신 해당 PTE에 저장한다.)
- Lesson:efficient footprint caching for tagless dram caches 9317d60c + (F-TDC는 cached page footprint를 cache-tag array가 아니라 TLB에서 추적하고 eviction 시 footprint history를 제한된 on-chip table 대신 해당 PTE에 저장한다.)
- Lesson:sieve is simpler than lru an efficient turn key eviction algorithm for web caches 86f86293 + (FIFO 큐, hand, lazy promotion, quick demotion만으로 SIEVE를 구성한다.)
- Lesson:faasmem improving memory efficiency of serverless computing with memory pool architecture b6ab9eec + (FaaSMem groups pages into Page Buckets by allocation stage, offloads segments selectively to a memory pool, and keeps semi-warm functions alive for a bounded period.)
- Lesson:fetchbpf customizable prefetching policies in linux with ebpf 9756cb90 + (FetchBPF exposes memory-prefetch hooks and policy state through eBPF, enabling verified user-defined policies in a stock-kernel framework.)
- Lesson:optimizing file systems on heterogeneous memory by integrating dram cache with virtual memory ma aaa7aa17 + (FlacFS는 VM에 DRAM 캐시를 통합해 zero-copy caching과 병렬 관리를 제공하는 library FS다.)
- Lesson:fleetio managing multi tenant cloud storage with multi agent reinforcement learning 714ca9fb + (FleetIO uses multi-agent RL integrated with storage scheduling, clusters runtime traces into latency- or bandwidth-oriented types, and tunes rewards for fine-grained harvesting.)
- Lesson:flexmem adaptive page profiling and migration for tiered memory 43991e2a + (FlexMem combines performance counters with hint faults, dynamically adjusts demotion volume, and tracks warm-page ranges.)
- Lesson:technical review request aware cooperative i o scheduling for scale out database applications fe4a3e09 + (Front-end에서 request arrival order를 ID로 포착해 back-end I/O까지 전달하고, 각 노드가 같은 global request order를 존중하도록 cooperative scheduler를 구현한다.)
- Lesson:request aware cooperative i o scheduling for scale out database applications 25327d2c + (Front-end에서 request arrival order를 ID로 포착해 back-end I/O까지 전달하고, 각 노드가 같은 global request order를 존중하도록 cooperative scheduler를 구현한다.)
- Lesson:inf2 high throughput generative inference of large language models using near storage processing 3fcead73 + (HILOS는 상용 Samsung SmartSSD에서 구현한 PyTorch 통 … HILOS는 상용 Samsung SmartSSD에서 구현한 PyTorch 통합 near-storage processing framework입니다. 각 SSD의 FPGA accelerator가 decoding 단계의 exact attention을 처리하고, 큰 historical KV cache는 device-local 경로에 둡니다. Host interconnect에는 attention 입력과 최종 출력만 보내며, batch와 attention-head 차원으로 여러 SmartSSD에 작업을 분할합니다.</br></br>Cooperative X-cache는 선택 구간의 K와 V를 모두 보관하는 대신 절반 크기의 pre-projection activation X를 GPU로 읽어 K/V를 재생성합니다. Near-storage accelerator가 나머지 KV를 처리하는 동안 유휴 GPU와 GPUDirect Storage를 병렬 사용하며, 분석 모델로 비율을 정합니다.</br></br>Delayed KV-cache writeback은 head별 약 256-byte 신규 KV entry를 host memory에 모읍니다. CPU가 buffered key의 query-key partial dot product를 미리 계산하고, accelerator에는 scalar와 새 value만 보냅니다. 16개 entry가 4-KiB SSD page를 이룰 때 비동기로 spill하여 small write와 write amplification을 critical path에서 제거합니다.</br></br>FPGA attention accelerator는 sequence 길이에 비례하는 on-chip buffer 없이 block-temporal pipeline으로 exact attention을 수행합니다. 2-pass numerically stable softmax, online block transpose, GQA broadcasting, FP16 저장과 FP32 누산·지수 연산, 128 MAC units, storage-read/compute overlap을 사용합니다. HLS customization flow, C++ middleware, pybind11 PyTorch binding과 asynchronous writeback manager를 포함합니다.ng과 asynchronous writeback manager를 포함합니다.)
- Lesson:technical review fully harnessing the performance potential of dram less mobile flash storage d4449ca9 + (HP-UFS는 host driver와 FTL firmware가 수정 L2P mapping, valid-page bitmap/count를 host DRAM에 최신 상태로 유지하도록 협업한다. LRU 캐시, mapping writeback, crash-recovery 절차를 두고 UFS command/response의 빈 필드로 메타데이터를 piggyback한다.)
- Lesson:fully harnessing the performance potential of dram less mobile flash storage ff995292 + (HP-UFS는 host driver와 FTL firmware가 수정 L2P mapping, valid-page bitmap/count를 host DRAM에 최신 상태로 유지하도록 협업한다. LRU 캐시, mapping writeback, crash-recovery 절차를 두고 UFS command/response의 빈 필드로 메타데이터를 piggyback한다.)
- Lesson:hotrap hot record retention and promotion for lsm trees with tiered storage 849aa350 + (HotRAP adds a fine-grained on-disk hot-key tracker and two pathways for promoting selected records into RocksDB's faster levels.)
- Lesson:scalable and effective page table and tlb management on numa systems 92f46a49 + (Hydra는 투명한 on-demand partial replication과 정확한 sharer 추적으로 불필요한 TLB shootdown을 줄인다.)
- Lesson:ic cache efficient large language model serving via in context caching 4f05d8b7 + (IC-Cache selects historical Q/A examples, routes requests adaptively, and replays them with a cost-aware policy.)
- Lesson:ice collaborating memory and process management for user experience on resource limited mobile d 312e453f + (ICE identifies background processes likely to cause frequent refaults, freezes them, and thaws them when memory conditions permit.)
- Lesson:impress an importance informed multi tier prefix kv storage system for large language model infe 51b47015 + (IMPRESS exploits similarity in important-token indices across attention heads to identify and load only important KVs, then manages prefix state across storage/cache tiers by importance.)
- Lesson:technical review a neural network accelerator for mobile application processors d4383cb5 + (Inter-neuron과 intra-neuron parallelism을 모두 활용할 수 있는 mobile neural-network accelerator를 설계한다.)
- Lesson:a neural network accelerator for mobile application processors 72ce8850 + (Inter-neuron과 intra-neuron parallelism을 모두 활용할 수 있는 mobile neural-network accelerator를 설계한다.)
- Lesson:memory deduplication in mobile systems 063e8b2e + (KAIST 연구실과 연세대 연구실의 출판 목록을 확인했지만 poster PDF, 공식 초록, DOI와 artifact를 찾지 못했다. 중복 탐지·병합 방식과 평가 절차는 미확인이다.)
- Lesson:fast flexible and practical kernel extensions 9eb8897f + (KFlex separates kernel-resource safety from extension-resource safety, using automated verification plus lightweight runtime checks while remaining backward-compatible with eBPF.)
- Lesson:gqa training generalized multi query transformer models from multi head checkpoints a6839ee2 + (KV head를 중간 개수의 그룹으로 묶는 GQA와 MHA 체크포인트 업트레이닝을 제안한다.)
- Lesson:kvaccel a novel write accelerator for lsm tree based kv stores with host ssd collaboration 968f5966 + (KVACCEL partitions NAND between block and KV interfaces, uses the KV region as a temporary write buffer, and adds an in-device LSM buffer plus range-scan iterator.)