USENIX ATC '21
2021-07-14..2021-07-16
-- online --
as the organisers put it
2021 USENIX Annual Technical Conference will take place as a virtual event on July 14–16, 2021. USENIX ATC '21 will bring together leading systems researchers for cutting-edge systems research and the opportunity to gain insight into a wealth of must-know topics.
| title | author | length |
|---|---|---|
| Distributed Trust: Is “Blockchain” the answer? | Radia Perlman | 54m |
Claims that blockchain gives universal Byzantine robustness fail against concrete issues: forks from network partitions or greater compute power, so… | ||
| Naos: Serialization-free RDMA networking in Java | Konstantin Taranov, Rodrigo Bruno, Gustavo Alonso, Torsten Hoefler | 15m |
Naos implements a serialization-free Java networking layer that uses RDMA write to remote heap to copy on-heap objects verbatim, then runs a DFS-based… | ||
| One-sided RDMA-Conscious Extendible Hashing for Disaggregated Memory | Pengfei Zuo, Jiazhao Sun, Liu Yang, Shuangwu Zhang, Yu Hua | 15m |
RACE hashing is an RDMA-conscious extendible hash table that executes every index operation using one-sided RDMA verbs, avoids remote compute by keeping all… | ||
| Characterizing and Optimizing Remote Persistent Memory with RDMA and NVM | Xingda Wei, Xiating Xie, Rong Chen, Haibo Chen, Binyu Zang | 12m |
Systematically measures RDMA accessing real production NVM (Intel 3D XPoint) and extracts nine concrete optimization hints across configuration (e.g. disable… | ||
| MigrOS: Transparent Live-Migration Support for Containerised RDMA Applications | Maksym Planeta, Jan Bierbaum, Leo Sahaya Daphne Antony, Torsten Hoefler, Hermann Härtig | 15m |
MigrOS provides transparent live migration for containerised RDMA applications by making minimal changes to the RoCE v2 protocol and kernel RDMA stack so user… | ||
| Prediction-Based Power Oversubscription in Cloud Platforms | Alok Gautam Kumbhare, Reza Azimi, Ioannis Manousakis, Anand Bonde, Felipe Vieira Frujeri, Nithish Mahalingam, Pulkit A. Misra, Seyyed Ahmad Javadi, Bianca Schroeder, Marcus Fontoura, Ricardo Bianchini | 16m |
Claims a 2x increase in safe power oversubscription by combining an ML model in Resource Central (99% accuracy at identifying user‑facing VMs) with enhanced VM… | ||
| Proactive Energy-Aware Adaptive Video Streaming on Mobile Devices | Jiayi Meng, Qiang Xu, Y. Charlie Hu | 15m |
Integrates a user energy budget into ABR logic by adding a power predictor and controller to an MPC-based ABR with power predictor, implemented for 360° video… | ||
| Video Analytics with Zero-streaming Cameras | Mengwei Xu, Tiantu Xu, Yunxin Liu, Felix Xiaozhu Lin | 15m |
DIVA is a camera/cloud runtime that leaves video on local SD cards and answers queries by (1) building sparse landmark frames (e.g., one every ~30s) that… | ||
| ASAP: Fast Mobile Application Switch via Adaptive Prepaging | Sam Son, Seung Yul Lee, Yunho Jin, Jonghyun Bae, Jinkyu Jeong, Tae Jun Ham, Jae W. Lee, Hongil Yoon | 15m |
ASAP implements adaptive prepaging inside Android with a switch footprint estimator (static profiles for file-backed pages, dynamic tables for anonymous pages)… | ||
| PYLIVE: On-the-Fly Code Change for Python-based Online Services | Haochen Huang, Chengcheng Xiang, Li Zhong, Yuanyuan Zhou | 14m |
A framework that enables on-the-fly code change for Python online services, using Python’s meta-object protocol and dynamic typing to add dynamic logging and… | ||
| RIFF: Reduced Instruction Footprint for Coverage-Guided Fuzzing | Mingzhe Wang, Jie Liang, Chijin Zhou, Yu Jiang, Rui Wang, Chengnian Sun, Jiaguang Sun | 14m |
RIFF reduces fuzzing overhead by moving runtime computations to instrumentation-time and by turning per-block coverage writes into single-instruction… | ||
| TCP-Fuzz: Detecting Memory and Semantic Bugs in TCP Stacks with Fuzzing | Yong-Hao Zou, Jia-Ju Bai, Jielong Zhou, Jianfeng Tan, Chenggang Qin, Shi-Min Hu | 13m |
A fuzzing framework named TCP-Fuzz is built to test TCP stacks using dependency-based mutations for packet and syscall sequences, a branch transition metric… | ||
| MLEE: Effective Detection of Memory Leaks on Early-Exit Paths in OS Kernels | Wenwen Wang | 15m |
A static detector called MLEE analyzes compiled RVM IR of Linux 5.0, identifies conditional early-exit branches, cross-checks allocations versus deallocations… | ||
| Argus: Debugging Performance Issues in Modern Desktop Applications... | Lingmei Weng, Peng Huang, Jason Nieh, Junfeng Yang | 15m |
Argus is a causal-tracing system that annotates trace graphs with strong and weak edges and uses a beam-search diagnosis to tolerate tracing ambiguities. It… | ||
| aDFS: An Almost Depth-First-Search Distributed Graph-Querying System | Vasileios Trigonakis, Jean-Pierre Lozi, Tomáš Faltín, Nicholas P. Roth, Iraklis Psaroudakis, Arnaud Delamare, Vlad Haprian, Călin Iorgulescu, Petr Koupy, Jinsoo Lee, Sungpack Hong, Hassan Chafi | 14m |
aDFS is a distributed graph query engine that executes pattern matching with an almost depth-first exploration, uses non-blocking dispatch of remote… | ||
| GLIST: Towards In-Storage Graph Learning | Cangyuan Li, Ying Wang, Cheng Liu, Shengwen Liang, Huawei Li, Xiaowei Li | 14m |
GLIST moves GNN computation into SSDs using an in-storage FPGA accelerator plus an embedded CPU, flash arrays and DRAM, implements sampler / aggregate /… | ||
| DART: A Scalable and Adaptive Edge Stream Processing Engine | Pinchao Liu, Dilma Da Silva, Liting Hu | 16m |
DART implements a fully decentralized edge stream engine that uses a DHT-based overlay (Pastry) and runs on Apache Flume to place, chain and scale operators… | ||
| CrystalPerf: Learning to Characterize the Performance of Dataflow Computation... | Huangshi Tian, Minchen Yu, Wei Wang | 14m |
CrystalPerf learns from source code to infer per-operator resource breakdowns (CPU, network, disk) and combines that with non‑intrusive sampling profiler… | ||
| Controlling Memory Footprint of Stateful Streaming Graph Processing | Pourya Vaziri, Keval Vora | 16m |
Two memory-saving iterative models are implemented inside graph code: the Selective Stateful Iterative Model keeps intermediate state for a tunable subset… | ||
| Avocado: A Secure In-Memory Distributed Storage System | Maurice Bailleu, Dimitra Giantsidi, Vasilis Gavrielatos, Do Le Quoc, Vijay Nagarajan, Pramod Bhatotia | 14m |
Avocado is a secure in‑memory distributed key‑value store that uses Intel SGX to run the replication protocol inside enclaves, splits keys (in‑enclave) and… | ||
| Accelerating Encrypted Deduplication via SGX | Yanjing Ren, Jingwei Li, Zuoru Yang, Patrick P. C. Lee, Xiaosong Zhang | 15m |
System implements SGXDedup: an SGX-enclave based, server-aided MLE system that moves key generation and proof-of-ownership into enclaves. Claims 131.9× faster… | ||
| ICARUS: Attacking low Earth orbit satellite networks | Giacomo Giuliari, Tommaso Ciussani, Adrian Perrig, Ankit Singla | 15m |
ICARUS shows a practical denial‑of‑service on LEO constellations by exploiting public orbital positions, predictable routing, and global terminal access to… | ||
| RainBlock: Faster Transaction Processing in Public Blockchains | Soujanya Ponnapalli, Aashaka Shah, Souvik Banerjee, Dahlia Malkhi, Amy Tai, Vijay Chidambaram, Michael Wei | 16m |
RainBlock removes the I/O bottleneck in public proof-of-work chains by decoupling state into sharded storage nodes, adding io helpers that prefetch compact… | ||
| An Off-The-Chain Execution Environment for Scalable Testing and Profiling... | Yeonsoo Kim, Seongho Jeong, Kamil Jezek, Bernd Burgstaller, Bernhard Scholz | 14m |
An off‑chain replayer stores per-transaction input-output tuples in compact substates and replays 590 million historical transactions from the first 9M blocks… | ||
| Octo: INT8 Training with Loss-aware Compensation and Backward Quantization for Tiny | Qihua Zhou, Song Guo, Zhihao Qu, Jingcai Guo, Zhenda Xu, Jiewei Zhang, Tao Guo, Boyuan Luo, Jingren Zhou | 14m |
Octo implements end-to-end INT8 training using a new Loss-aware Compensation mechanism and Parameterized Range Clipping so both forward and backward dot… | ||
| Fine-tuning giant neural networks on commodity hardware with automatic pipeline... | Saar Eliad, Ido Hakimi, Alon De Jagger, Mark Silberstein, Assaf Schuster | 14m |
FTPipe presents a practical pipeline model‑parallel system that fine‑tunes giant transformers (BERT‑340M, GPT2‑1.5B, T5‑3B) on commodity PCIe GPUs by… | ||
| INFaaS: Automated Model-less Inference Serving | Francisco Romero, Qian Li, Neeraja J. Yadwadkar, Christos Kozyrakis | 15m |
INFaaS automatically generates and profiles hundreds of model variants from trained models, then picks per-query combinations of model, framework, and hardware… | ||
| Jump-Starting Multivariate Time Series Anomaly Detection for Online Service Systems | Minghua Ma, Shenglin Zhang, Junjie Chen, Jim Xu, Haozhe Li, Yongliang Lin, Xiaohui Nie, Bo Zhou, Yong Wang, Dan Pei | 14m |
JumpStarter applies Compressed Sensing with shape-based clustering and outlier-resistant sampling to multivariate time series, reconstructing clustered metric… | ||
| Palleon: A Runtime System for Efficient Video Processing toward Dynamic Class Skew | Boyuan Feng, Yuke Wang, Gushu Li, Yuan Xie, Yufei Ding | 13m |
Palleon is a runtime system that detects dynamic class skew from prediction streams, uses early sampling and windowed switch detection, applies a Bayesian… | ||
| FaaSNet: Scalable and Fast Provisioning of Custom Serverless Container Runtimes... | Ao Wang, Shuai Chang, Huangshi Tian, Hongqi Wang, Haoran Yang, Huiba Li, Rui Du, Yue Cheng | 12m |
FaaSNet implements a lightweight, hierarchical function tree plus an I/O‑efficient on‑demand fetching protocol inside Alibaba Cloud Function Compute to cut… | ||
| Experiences in Managing the Performance and Reliability of a Large-Scale Genomics... | Michael Hao Tong, Robert L. Grossman, Haryadi S. Gunawi | 15m |
GDC (Genomic Data Commons) pipelines on KVM VMs age over days, producing rising EPT violations that correlate with multi× slowdown; repeated runs (five… | ||
| Scaling Large Production Clusters with Partitioned Synchronization | Yihui Feng, Zhi Liu, Yunjian Zhao, Tatiana Jin, Yidi Wu, Yang Zhang, James Cheng, Chao Li, Tao Guan | 14m |
Clusters run 100,000 machines and process 3.1–4.4 billion tasks (3.3–4.4 million jobs) per month with 87% of tasks under 10s, so the scheduler is redesigned to… | ||
| Fighting the Fog of War: Automated Incident Detection for Cloud Systems | Liqun Li, Xu Zhang, Xin Zhao, Hongyu Zhang, Yu Kang, Pu Zhao, Bo Qiao, Shilin He, Pochian Lee, Jeffrey Sun, Feng Gao, li yang, Qingwei Lin, Saravanakumar Rajmohan, Zhangwei Xu, Dongmei Zhang | 15m |
Warden is an automated incident detector deployed in Azure IcM that ingests alerts from selected monitors, uses a balanced random forest over 3-hour sliding… | ||
| Habitat: A Runtime-Based Computational Performance Predictor... | Geoffrey X. Yu, Yubo Gao, Pavel Golikov, Gennady Pekhimenko | 16m |
Habitat predicts a single training-iteration execution time by profiling operations on an available GPU and scaling per-kernel timings to other GPUs, using… | ||
| Zico: Efficient GPU Memory Sharing for Concurrent DNN Training | Gangmuk Lim, Jeongseob Ahn, Wencong Xiao, Youngjin Kwon, Myeongjae Jeon | 16m |
Zico builds a shared GPU memory pool plus a memory manager that uses CUDA events and an in‑flight kernel counter to safely reclaim and limit allocations, and a… | ||
| Refurbish Your Training Data: Reusing Partially Augmented Samples... | Gyewon Lee, Irene Lee, Hyeonmin Ha, Kyunggeun Lee, Hwarim Hyun, Ahnjae Shin, Byung-Gon Chun | 15m |
Claims a pipeline change that splits augmentation into partial augmentation and final augmentation and caches partially augmented samples to cut CPU overhead… | ||
| ZeRO-Offload: Democratizing Billion-Scale Model Training | Jie Ren, Samyam Rajbhandari, Reza Yazdani Aminabadi, Olatunji Ruwase, Shuangyan Yang, Minjia Zhang, Dong Li, Yuxiong He | 14m |
ZeRO-Offload offloads fp16 gradients and fp32 optimizer states and runs parameter updates on CPU while keeping forward/backward on GPU, enabling 30 billion… | ||
| Hashing Linearity Enables Relative Path Control in Data Centers | Zhehui Zhang, Haiyang Zheng, Jiayao Hu, Xiangning Yu, Chenchen Qi, Xuemei Shi, Guohui Wang | 13m |
Reveals that ECMP hashing in real switch ASICs uses hashing linearity (XOR/CRC) so flipping header bits gives deterministic path offsets, and implements RePaC… | ||
| Live in the Express Lane | Patrick Jahnke, Vincent Riesop, Pierre-Louis Roman, Pavel Chuprikov, Patrick Eugster | 13m |
X-Lane isolates coordination traffic on a dedicated core and NIC queue, implemented as a kernel module and offloaded to a Metronome NFP4000 smartNIC, yielding… | ||
| Understanding Precision Time Protocol in Today's Wi-Fi Networks: A Measurement Study | Paizhuo Chen, Zhice Yang | 13m |
USENIX ATC '21 - Understanding Precision Time Protocol in Today's Wi-Fi Networks: A Measurement Study
Paizhuo Chen and Zhice Yang, ShanghaiTech… | ||
| AUTO: Adaptive Congestion Control Based on Multi-Objective Reinforcement Learning... | Xu Li, Feilong Tang, Jiacheng Liu, Laurence T. Yang, Luoyi Fu, Long Chen | 13m |
AUTO formulates congestion control as a multi-objective MDP and trains a MORL agent plus a preference adaptation model to produce the Pareto frontier of… | ||
| Hey, Lumi! Using Natural Language for Intent-Based Network Management | Arthur S. Jacobs, Ricardo J. Pfitscher, Rafael H. Ribeiro, Ronaldo A. Ferreira, Lisandro Z. Granville, Walter Willinger, Sanjay G. Rao | 13m |
Builds a working system called Lumi that turns freeform operator requests into a structured intent language (Nile), compiles them into device configs (targets… | ||
| Boosting Full-Node Repair in Erasure-Coded Storage | Shiyao Lin, Guowen Gong, Zhirong Shen, Patrick P. C. Lee, Jiwu Shu | 16m |
RepairBoost is a scheduling framework that speeds full-node repair for linear erasure codes by building a DAG repair abstraction, balancing upload and download… | ||
| KVIMR: Key-Value Store Aware Data Management Middleware for Interlaced Magnetic... | Yuhong Liang, Tsun-Yu Yang, Ming-Chang Yang | 15m |
KVIMR is a middleware that sits between LSM-tree KV stores (RocksDB and others) and emulated 100 GB IMR HDD to cut costly remote rewrites. It implements… | ||
| Differentiated Key-Value Storage Management for Balanced I/O Performance | Yongkun Li, Zhen Liu, Patrick P. C. Lee, Jiayu Wu, Yinlong Xu, Yi Wu, Liu Tang, Qi Liu, Qiu Cui | 13m |
DiffKV builds a prototype on TiKV/Titan that splits keys and values and uses a V3 multi-level value tree plus fine-grained KV separation (small/medium/large)… | ||
| ZNS: Avoiding the Block Interface Tax for Flash-based SSDs | Matias Bjørling, Abutalib Aghayev, Hans Holmberg, Aravind Ramesh, Damien Le Moal, Gregory R. Ganger, George Amvrosiadis | 16m |
ZNS exposes erase-block boundaries and append-only zones so host software moves fine-grained placement into userspace; implemented Linux NVMe+block changes… | ||
| MapperX: Adaptive Metadata Maintenance for Fast Crash Recovery of DM-Cache Based... | Lujia Yin, Li Wang, Yiming Zhang, Yuxing Peng | 15m |
Maps dirty-bit metadata into an on-disk adaptive bit-tree (ABT) that is synchronously updated per write but adapts its leaf granularity to exploit spatial… | ||
| AI in Finance: Scope and Examples | Manuela Veloso | 59m |
Describes an image-first pipeline that converts S&P500 time series into pictures, trains a Mondrian image classifier for buy/no‑buy and an autoencoder that… | ||
| It's Time for Operating Systems to Rediscover Hardware | Timothy Roscoe | 01h 06m |
Argues that modern hardware is wildly heterogeneous (SoCs with many processors, nonuniform address spaces) and that Linux only controls a small cluster while… | ||
| Exploring the Design Space of Page Management for Multi-Tiered Memory Systems | Jonghyeon Kim, Wonkyo Choe, Jeongseob Ahn | 15m |
Linux 5.3 is modified and evaluated on a two-socket system combining DRAM and Intel DCPMM to expose page placement choices. The design implements… | ||
| A Fast and Flexible Hardware-based Virtualization Mechanism for Computational... | Dongup Kwon, Dongryeong Kim, Junehyuk Boo, Wonsik Lee, Jangwoo Kim | 13m |
An FPGA card called FCSV-Engine implements hardware‑assisted virtualization for both the computation and storage blocks, extends NVMe with a new command… | ||
| Fair Scheduling for AVX2 and AVX-512 Workloads | Mathias Gottschlag, Philipp Machauer, Yussuf Khalil, Frank Bellosa | 15m |
Modern Intel CPUs drop frequency for AVX2/AVX‑512 code and that slowdown leaks to co‑running tasks. The paper measures a 24.9% average slowdown of non‑AVX work… | ||
| SKQ: Event Scheduling for Optimizing Tail Latency in a Traditional OS Kernel | Siyao Zhao, Haoyu Gu, Ali José Mashtizadeh | 15m |
Schedulable Kqueue (SKQ) replaces FreeBSD kqueue internals with a scalable architecture and event scheduler that routes events to per-thread queues and… | ||
| A Linux Kernel Implementation of the Homa Transport Protocol | John Ousterhout | 15m |
Homa/Linux is a Homa/Linux kernel module that implements a message oriented transport and a message-oriented RPC API with three syscalls and runs on unmodified… | ||
| Ayudante: A Deep Reinforcement Learning Approach to Assist Persistent Memory... | Hanxian Huang, Zixuan Wang, Juno Kim, Steven Swanson, Jishen Zhao | 15m |
A deep RL model translates C/C++/Java volatile code into PM-aware code by inserting PMDK annotations, then a three-level code refining pipeline runs gcc, PM… | ||
| TIPS: Making Volatile Index Structures Persistent with DRAM-NVMM Tiering | R. Madhava Krishnan, Wook-Hee Kim, Xinwei Fu, Sumit Kumar Monga, Hee Won Lee, Minsung Jang, Ajit Mathew, Changwoo Min | 15m |
TIPS is a framework that makes existing volatile index structures persistent by using DRAM-NVMM tiering and a UNO logging hybrid low-overhead log to guarantee… | ||
| Improving Performance of Flash Based Key-Value Stores Using Storage Class Memory... | Hiwot Tadese Kassa, Jason Akers, Mrinmoy Ghosh, Zhichao Cao, Vaibhav Gogte, Ronald Dreslinski | 15m |
Implementation uses Intel Optane DCPMM in AppDirect mode with memkind to extend memory and a RocksDB hybrid cache that applies a dram-first admission policy… | ||
| First Responder: Persistent Memory Simultaneously as High Performance Buffer... | Hyunsub Song, Shean Kim, J. Hyun Kim, Ethan JH Park, Sam H. Noh | 15m |
Build: a PM-resident cache/storage layer inside the block I/O path that keeps chunks + tags with a two-way layout, uses static placement and periodic… | ||
| A Case Study of Processing-in-Memory in off-the-Shelf Systems | Joel Nider, Craig Mustard, Andrada Zoltan, John Ramsden, Larry Liu, Jacob Grossbard, Mohammad Dashti, Romaric Jodin, Alexandre Ghiti, Jordi Chauzi, Alexandra Fedorova | 12m |
Evaluation of UPMEM PIM on a real server shows internal memory bandwidth and compute scale linearly with added DPUs, but practical limits appear: chip… | ||
| XFUSE: An Infrastructure for Running Filesystem Services in User Space | Qianbo Huai, Windsor Hsu, Jiwei Lu, Hao Liang, Haobo Xu, Wei Chen | 12m |
XFUSE is a FUSE-compatible framework that reduces user-space filesystem latency to the 3–4 microsecond range and demonstrates >8 GB/s throughput by adding an… | ||
| Max: A Multicore-Accelerated File System for Flash Storage | Xiaojian Liao, Youyou Lu, Erci Xu, Jiwu Shu | 15m |
Max implements a multicore-friendly log-structured file system in the Linux kernel (based on F2FS) and demonstrates concrete speedups: evaluated on a 72-core… | ||
| Z-Journal: Scalable Per-Core Journaling | Jongseok Kim, Cassiano Campes, Joo-Young Hwang, Jinkyu Jeong, Euiseong Seo | 15m |
Z-Journal implements per-core journal stacks in the Linux kernel (replacing JBD2) and modifies ext4 to record a system-wide transaction order graph via… | ||
| LODIC: Logical Distributed Counting for Scalable File Access | Jeoungahn Park, Taeho Hwang, Jongmoo Choi, Changwoo Min, Youjip Won | 15m |
A Linux kernel patch implements a per-process reference counter that embeds small local counters into unused bits of x86 PTEs and locates them by scanning a… | ||
| UniStore: A fault-tolerant marriage of causal and strong consistency | Manuel Bravo, Alexey Gotsman, Borja de Régil, Hengfeng Wei | 14m |
UniStore is a transactional POR store that actually implements optimistic certification using two‑phase commit + Paxos and a lightweight uniformity check that… | ||
| Optimistic Concurrency Control for Real-world Go Programs | Zhizhou Zhang, Milind Chabbi, Adam Welc, Timothy Sherwood | 12m |
Gocc is a source-to-source transformer that replaces mutex-protected regions in Go with Hardware Transactional Memory using inter-procedural AST analysis to… | ||
| Faastlane: Accelerating Function-as-a-Service Workflows | Swaroop Kotni, Ajay Nayak, Vinod Ganapathy, Arkaprava Basu | 15m |
Faastlane runs workflow functions as threads within a single process (on Apache OpenWhisk) so intermediate data travels by loads/stores instead of network or… | ||
| SONIC: Application-aware Data Passing for Chained Serverless Applications | Ashraf Mahgoub, Karthick Shankar, Subrata Mitra, Ana Klimovic, Somali Chaterji, Saurabh Bagchi | 16m |
Compares three concrete data-passing modes VM-Storage, Direct-Passing, Remote-Storage and shows no single mode wins: small fan-out favors VM-Storage, medium… | ||
| nothing matches | ||