FAST '25
as the organisers put it
23rd USENIX Conference on File and Storage Technologies (FAST '25), February 25–27, 2025, Hyatt Regency Santa Clara, Santa Clara, CA, USA. FAST brings together storage-system researchers and practitioners to explore new directions in the design, implementation, evaluation, and deployment of storage systems.
| title | author | length |
|---|---|---|
| Keynote Address: Insights Gained from Delivering Two Generations of AI Supercomputers... | Seetharami Seelam | 52m |
IBM Cloud deploys two generations of Vela systems (2022 → expanded 2023/2024) that virtualize GPUs with ~90% efficiency, move from TCP to GPU direct RDMA over… | ||
| Fast, Transparent Filesystem Microkernel Recovery with Ananke | Jing Liu, Yifan Dai, Andrea C. Arpaci-Dusseau, Remzi H. Arpaci-Dusseau | 16m |
Ananke is a filesystem microkernel service that records an in-memory plog and runs an M algorithm to decide which operations to ignore, modify, or replay, then… | ||
| Boosting File Systems Elegantly: A Transparent NVM Write-ahead Log for Disk File Systems | Guoyu Wang, Xilong Che, Haoyang Wei, Shuo Chen, Puyi He, Juncheng Hu | 18m |
NVLog is an NVM write‑ahead log that intercepts VFS syncs to persist only synchronous bytes with byte-granularity sync absorption, keeping asynchronous traffic… | ||
| DJFS : Directory-Granularity Filesystem Journaling for CMM-H SSDs | Seung Won Yoo, Joontaek Oh, Myeongin Cheon, Bonmoo Koo, Wonseb Jeong, Hyunsub Song, Hyeonho Song, Donghun Lee, Youjip Won | 16m |
DJFS implements per-directory transactions for CXL.ME (CMH) SSDs, using path-based transaction selection, transaction coalescing, and transaction conflict… | ||
| ScaleLFS: A Log-Structured File System with Scalable Garbage Collection for Commodity/// | Jin Yong Ha, Sangjin Lee, Hyeonsang Eom, Yongseok Son | 16m |
ScaleLFS implements per-core dedicated GC, a scalable victim manager using atomic bitmaps and loose-sync counters, and a scalable victim protector that enables… | ||
| Rethinking the Request-to-IO Transformation Process of File Systems for Full... | Yekang Zhan, Haichuan Hu, Xiangrui Yang, Qiang Cao, Hong Jiang, Shaohua Wang, Jie Yao | 16m |
Experimental analysis shows file systems pay a ~9× penalty when fsync exposes page-cache work and small SSD reads cause up to an 80% latency increase for 1 MB… | ||
| FlacIO: Flat and Collective I/O for Container Image Service | Yubo Liu, Hongbo Li, Mingrui Liu, Rui Jing, Jian Guo, Bo Zhang, Hanjun Guo, Yuxin Ren, Ning Jia | 17m |
FlacIO implements a new runtime image abstraction plus a kernel runtime page cache and a userland FlacIO driver to replace storage-oriented images with… | ||
| Cloudscape: A Study of Storage Services in Modern Cloud Architectures | Sambhav Satija, Chenhao Ye, Ranjitha Kosgi, Aditya Jain, Romit Kankaria, Yiwei Chen, Andrea C. Arpaci-Dusseau, Remzi H. Arpaci-Dusseau, Kiran Srinivasan | 17m |
Cloudscape converts ~400 AWS “This is my architecture” videos into structured graphs via iterative collaborative coding and manual annotation and ships a… | ||
| Maat: Analyzing and Optimizing Overcharge on Blockchain Storage | Zheyuan He, Zihao Li, Ao Qiao, Jingwei Li, Feng Luo, Sen Yang, Gelei Deng, Shuwei Song, Xiaosong Zhang, Ting Chen, Xiapu Luo | 14m |
Maat analyzes three concrete overcharging patterns that treat memory reads/writes or duplicate contract writes as disk IO and it instruments storage-level gas… | ||
| Revisiting Network Coding for Warm Blob Storage | Chuang Gan, Yuchong Hu, Leyan Zhao, Xin Zhao, Pengyu Gong, Dan Feng | 10m |
NCBlob is a C++ Linux prototype that encodes small blobs with non-systematic MSR codes and large blobs with systematic MSR, adds a locality manager, and uses… | ||
| Mooncake: Trading More Storage for Less Computation — A KVCache-centric Architecture... | Ruoyu Qin, Zheming Li, Weiran He, Jialei Cui, Feng Ren, Mingxing Zhang, Yongwei Wu, Weimin Zheng, Xinran Xu | 17m |
Mooncake builds a KVCache-centric global cache and disaggregates prefill and decode so prefill computes once and decoding pulls reusable KVCache blocks from… | ||
| Towards High-throughput and Low-latency Billion-scale Vector Search via CPU/GPU... | Bing Tian, Haikun Liu, Yuhang Tang, Shihai Xiao, Zhuohui Duan, Xiaofei Liao, Hai Jin, Xuecang Zhang, Junhua Zhu, Yu Zhang | 16m |
FusionANNS is a working SSD + single entry‑level GPU ANNS that uses CPU/GPU collaborative filtering to keep PQ‑compressed vectors in GPU memory, avoid… | ||
| IMPRESS: An Importance-Informed Multi-Tier Prefix KV Storage System for Large Language... | Weijian Chen, Shuibing He, Haoyang Qu, Ruidong Zhang, Siling Yang, Ping Chen, Yi Zheng, Baoxing Huai, Gang Chen | 13m |
IMPRESS is a storage stack that cuts time-to-first-token by up to 2.8× versus state-of-the-art while holding inference accuracy within ~0.2% of full… | ||
| GPHash: An Efficient Hash Index for GPU with Byte-Granularity Persistent Memory | Menglei Chen, Yu Hua, Zhangyu Chen, Ming Zhang, Gen Dong | 14m |
GPHash: An Efficient Hash Index for GPU with Byte-Granularity Persistent Memory
Menglei Chen, Yu Hua, Zhangyu Chen, Ming Zhang, and Gen Dong, Wuhan National… | ||
| GeminiFS: A Companion File System for GPUs | Shi Qiu, Weinan Liu, Yifan Hu, Jianqin Yan, Zhirong Shen, Xin Yao, Renhai Chen, Gong Zhang, Yiming Zhang | 18m |
GeminiFS embeds file metadata directly inside files so GPUs read address mappings without host roundtrips, extends the NVMe driver to map I/O queues into GPU… | ||
| 3L-Cache: Low Overhead and Precise Learning-based Eviction Policy for Caches | Wenbin Zhou, Zhixiong Niu, Yongqiang Xiong, Juan Fang, Qian Wang | 15m |
3L-Cache is an object-level learning eviction policy that uses GBM prediction, sampled training windows, bidirectional sampling, and heap+hash matching to cut… | ||
| LeapGNN: Accelerating Distributed GNN Training Leveraging Feature-Centric Model Migration | Weijian Chen, Shuibing He, Haoyang Qu, Xuechen Zhang | 14m |
Proposes feature-centric model migration that moves the (small) GNN model to servers holding vertex features, cutting remote feature transfers that dominate… | ||
| HiDPU: A DPU-Oriented Hybrid Indexing Scheme for Disaggregated Storage Systems | Wenbin Zhu, Zhaoyan Shen, Qian Wei, Renhai Chen, Xin Yao, Dongxiao Yu, Zili Shao | 18m |
HiDPU is a DPU-oriented hybrid index that partitions mapping entries into accurate, PTHash, and LPTHash segments, builds a layered learned index across segment… | ||
| PIMLex: A High-Performance Learned Index with Processing-in-Memory | Lixiao Cui, Kedi Yang, Yusen Li, Gang Wang, Xiaoguang Liu | 17m |
PIMLex uses a decoupled two-layer structure (small PIM-based search layer of anchor keys plus large DRAM data layer) and PIM-friendly choices: a lookup-table… | ||
| HaSiS: A Hardware-assisted Single-index Store for Hybrid Transactional and Analytical... | Kecheng Huang, Zhaoyan Shen, Zili Shao, Feng Chen, Tong Zhang | 15m |
A B+-tree single-index store leverages commercially available computational storage drives with transparent compression to run HTAP without dual indexes. The… | ||
| AegonKV: A High Bandwidth, Low Tail Latency, and Low Storage Cost KV-Separated LSM Store | Zhuohui Duan, Hao Feng, Haikun Liu, Xiaofei Liao, Hai Jin, Bangyu Li | 16m |
AegonKV moves garbage-collection for KV-separated LSM storage onto a Samsung first-generation SmartSSD-based GC offloading, implementing a bitmap valid map and… | ||
| D2FS: Device-Driven Filesystem Garbage Collection | Juwon Kim, Seungjae Lee, Joontaek Oh, Dongkun Shin, Youjip Won | 15m |
Claims that device-side GC can replace filesystem GC and cut overhead by letting the SSD reclaim invalid segments and push LBA remaps to the host. Implements… | ||
| ShiftLock: Mitigate One-sided RDMA Lock Contention via Handover | Jian Gao, Qing Wang, Jiwu Shu | 15m |
ShiftLock claims to cut RDMA lock retries by enabling direct client-to-client handovers so waiting clients spin locally instead of hammering the server; it… | ||
| Selective On-Device Execution of Data-Dependent Read I/O | Chanyoung Park, Minu Chung, Hyungon Moon | 15m |
Prototype runs atop the NVMeVirt storage emulator with an emulated ARM Cortex‑A53 on‑device CPU and Intel Optane DC as backing storage, and implements hybrid… | ||
| On Scalable Integrity Checking for Secure Cloud Disks | Quinn Burke, Ryan Sheatsley, Rachel King, Owen Hines, Michael Swift, Patrick McDaniel | 18m |
Paper quantifies storage-level Merkle tree costs and shows hashing is bottleneck, then designs Dynamic Merkle Trees that bias hot leaves toward the root… | ||
| Silhouette: Leveraging Consistency Mechanisms to Detect Bugs in Persistent Memory-Based.. | Bing Jiao, Ashvin Goel, An-I Andy Wang | 16m |
Silhouette is a bug‑detector for persistent‑memory file systems that uses static instrumentation and data-type-based dynamic analysis to verify… | ||
| OPIMQ: Order Preserving IO stack for Multi-Queue Block Device | Jieun Kim, Joontaek Oh, Juwon Kim, Seung Won Yoo, Youjip Won | 15m |
OPIMQ enforces storage order across host and multi‑queue devices by tagging each IO with stream and app IDs and by four concrete mechanisms: Epoch Pinning… | ||
| AWUPF Rediscovered: Atomic Writes to Unleash Pivotal Fault-Tolerance in SSDs | Jiyune Jeon, Jongseok Kim, Sam H. Noh, Euiseong Seo | 19m |
Claim: use the SSD NVMe AWUPF atomic-write unit (commonly 4KB AWUPF) to skip journaling for small Log‑RAID metadata updates, implemented in Poseidon OS by… | ||
| AtomicDisk: A Secure Virtual Disk for TEEs against Eviction Attacks | Hongliang Tian, Xinyi Yu, Shaowei Song, Qingsong Chen, Zhihao Zhang, Shiyu Wang, Weijie Liu, Erci Xu, Shoumeng Yan, Yiming Zhang | 17m |
SGX‑PFS uses Merkle hash trees, an in‑enclave cache and a recovery journal but allows privileged hosts to capture transient snapshots during cache eviction… | ||
| MedFS: Pursuing Low Update Overhead via Metadata-Enabled Delta Compression for... | Chao Wu, Cheng Ji, Li-Pin Chang, Zongwei Zhu, Congming Gao, Weichao Guo, Chao Yu, Wang Yanzhi | 18m |
MedFS implements metadata-aware delta compression on f2fs by embedding small XOR-derived deltas into unused inode inline space and spilling larger or evicted… | ||
| Don't Maintain Twice, It's Alright: Merged Metadata Management in Deduplication File... | Yanqi Pan, Wen Xia, Erci Xu, Hao Huang, Xiangyu Zou, Shiyi Li | 18m |
Deduplication metadata causes up to 38% I/O-path overhead; this work merges fingerprint-to-physical and L2P into a single on-disk entry so metadata persists… | ||
| Archer: Adaptive Memory Compression with Page-Association-Rule Awareness for High-Speed.. | Changlong Li, Zongwei Zhu, Chao Wang, Fangming Liu, Fei Xu, Edwin H. -M. Sha, Xuehai Zhou | 16m |
Archer implements an association-rule-aware memory compression framework integrated into the Linux kernel and ZRAM, with a prototype on Pixel and Huawei… | ||
| VectorCDC: Accelerating Data Deduplication with Vector Instructions | Sreeharsha Udayashankar, Abdelrahman Baba, Samer Al-Kiswany | 12m |
Presents a concrete method that rewrites hashless CDC algorithms into vector-friendly phases—tree-based extreme-byte search and packed range scans—implemented… | ||
| Oasis: An Out-of-core Approximate Graph System via All-Distances Sketches | Tsun-Yu Yang, Yi Li, Yizou Chen, Bingzhe Li, Ming-Chang Yang | 19m |
Oasis implements an out‑of‑core All‑Distances Sketch pipeline that builds ADS on storage using partitioning, equal‑size blocks and a duplicate active data… | ||
| PolyStore: Exploiting Combined Capabilities of Heterogeneous Storage | Yujie Ren, David Domingo, Jian Zhang, Paul John, Rekha Pitchumani, Sanidhya Kashyap, Sudarsun Kannan | 19m |
PolyStore is a meta layer that sits above device‑optimized file systems (Nova on persistent memory and ext4 on NVMe) and implements a user‑level runtime plus… | ||
| Liquid-State Drive: A Case for DNA Block Device for Enormous Data | Jiahao Zhou, Mingkai Dong, Fei Wang, Jingyao Zeng, Lei Zhao, Chunhai Fan, Haibo Chen | 13m |
LIQUID-STATE DRIVE implements a DNA-backed block device using small fast SSDs plus a simulator prototype and claims concrete reductions: write amplification… | ||
| DNA data storage: A generative tool for Motif-based DNA storage | Samira Brunmayr, Omer S. Sella, Thomas Heinis | 15m |
A generative algorithm builds motif alphabets base-by-base using Markov chains and log-score penalties for homopolymer, hairpins, GC content and to avoid key… | ||
| Honoring Erik Riedel | Garth A. Gibson | 06m |
Recognizes the loss of Erik Riedel and announces a memorial 6:30 p.m. tomorrow in the Elam room for reminiscing and family; outlines a career that spans 30… | ||
| Test of Time Award Presentation | Carl A. Waldspurger, Nohhyun Park, Alexander Garthwaite, Irfan Ahmad | 22m |
Efficient MRC construction uses deterministic spatial hashing to drive randomized sampling so full miss ratio curves are approximated from large traces without… | ||
| nothing matches | ||