USENIX ATC '22
2022-07-11..2022-07-13
Carlsbad, United States
| title | author | length |
|---|---|---|
| Surprise-Inspired Networking | David Tennenhouse | 50m |
Argues that information value comes from surprisal-driven fusion of edge-generated “new” data with cloud “prior” models, and proposes pushing computation and… | ||
| ZNSwap: un-Block your Swap | Shai Aviram Bergman, Niklas Cassel, Matias Bjørling, Mark Silberstein | 20m |
ZNSwap is a swap subsystem for Zoned Namespace SSDs that implements a space-efficient host-side GC and OS–device co‑ordination to avoid unnecessary internal… | ||
| Building a High-performance Fine-grained Deduplication Framework... | Xiangyu Zou, Wen Xia, Philip Shilane, Haijun Zhang, Haijun Zhang, Xuan Wang, Xuan Wang | 18m |
MeGA is a deduplication framework that applies selective data encoding to skip encoding chunks in containers rarely serving as bases, reorganizes storage into… | ||
| Secure and Lightweight Deduplicated Storage via Shielded Deduplication... | Zuoru Yang, Jingwei Li, Patrick P. C. Lee | 20m |
DEBE implements deduplication-before-encryption inside an Intel SGX enclave, using a count‑min sketch + top‑K heap to remove hot chunks in‑enclave and a… | ||
| uKharon: A Membership Service for Microsecond Applications | Rachid Guerraoui, Antoine Murat, Javier Picorel, Athanasios Xygkis, Huabing Yan, Pengfei Zuo | 19m |
uKharon is a membership service that uses RDMA to push membership and failover to microsecond scales: a multi-level failure detector (kernel crash… | ||
| KRCORE: A Microsecond-scale RDMA Control Plane for Elastic Computing | Xingda Wei, Fangming Lu, Rong Chen, Haibo Chen | 20m |
KRCORE is an RDMA library that creates full connections in 10μs connection establishment by virtualized kernel-space QPs and retrofitting dynamic connected… | ||
| Zero-Change Object Transmission for Distributed Big Data Analytics | Mingyu Wu, Shuaiwei Wang, Shuaiwei Wang, Haibo Chen, Binyu Zang | 20m |
Zero-Change Object Transmission (ZCOT) copies live Java objects into a shared exchange space mapped at the same addresses across JVMs and exposes a chunked… | ||
| Sift: Using Refinement-guided Automation to Verify Complex Distributed Systems | Haojun Ma, Hammad Ahmad, Aman Goel, Eli Goldweber, Jean-Baptiste Jeannin, Manos Kapritsos, Baris Kasikci | 20m |
Sift presents a concrete method that proves refinement between implementations and specs while automating most work: it uses refinement-guided automation plus… | ||
| Privbox: Faster System Calls Through Sandboxed Privileged Execution | Dmitry Kuznetsov, Adam Morrison | 22m |
Privbox implements a semi-privileged sandbox that runs annotated hot paths at ring 0 on Linux (LLVM toolchain) yet vets them with a verifier and… | ||
| BBQ: A Block-based Bounded Queue for Exchanging Data and Profiling | Jiawei Wang, Jiawei Wang, Diogo Behrens, Ming Fu, Lilith Oberhauser, Jonas Oberhauser, Jitang Lei, Geng Chen, Geng Chen, Hermann Härtig, Haibo Chen | 22m |
BBQ implements a block-based ringbuffer that splits the buffer into blocks so producers and consumers avoid enq‑deq contention when they operate on different… | ||
| Not that Simple: Email Delivery in the 21st Century | Florian Holzbauer, Johanna Ullrich, Olamide Omolola, Martina Lindorfer, Tobias Fiebig | 22m |
Study runs active experiments with eleven differently configured target domains and passive-DNS ranking to test real-world SMTP behavior. 622 participants from… | ||
| AddrMiner: A Comprehensive Global Active IPv6 Address Discovery System | Guanglei Song, Jiahai Yang, Lin He, Zhiliang Wang, Guo Li, Chenxin Duan, Yaozhong Liu, Zhongxiang Sun | 20m |
AddrMiner is a deployed IPv6 probing system that discovers 2.1 billion active IPv6 addresses over 13 months (1.7B de‑aliased, 0.4B aliased) and covers 86.4K… | ||
| Co-opting Linux Processes for High-Performance Network Simulation | Rob Jansen, Jim Newsome, Ryan Wails | 21m |
Phantom is a discrete‑event simulator that directly executes unmodified Linux processes via an LD_PRELOAD shim, traps noninterceptable syscalls with seccomp… | ||
| RunD: A Lightweight Secure Container Runtime for High-density Deployment... | Zijun Li, Jiagan Cheng, Quan Chen, Eryu Guan, Zizheng Bian, Yi Tao, Bin Zha, Qiang Wang, Weidong Han, Minyi Guo | 19m |
RunD is a secure-container runtime that splits rootfs with reflink-backed copy-on-write, ships a pre-patched guest kernel to maximize shareable text segments… | ||
| Help Rather Than Recycle: Alleviating Cold Startup in Serverless Computing... | Zijun Li, Linsong Guo, Quan Chen, Jiagan Cheng, Chuhao Xu, Deze Zeng, Zhuo Song, Tao Ma, Yong Yang, Chao Li, Minyi Guo | 18m |
System re-purposes idle warm containers from other functions instead of booting new ones, implemented as Pagurus with an intra-function manager, an… | ||
| RRC: Responsive Replicated Containers | Diyu Zhou, Yuval Tamir | 22m |
RRC implements hybrid checkpoint+replay replication at the container level and adds a new kernel primitive, container fork, that clones a container so… | ||
| Faith: An Efficient Framework for Transformer Verification on GPUs | Boyuan Feng, Tianqi Tang, Yuke Wang, Zhaodong Chen, Zhaodong Chen, Zheng Wang, Shu Yang, Shu Yang, Yuan Xie, Yufei Ding | 17m |
Faith implements a GPU framework that fuses bound-centric operations to cut redundant global memory accesses, loads weight once into shared memory, and… | ||
| DVABatch: Diversity-aware Multi-Entry Multi-Exit Batching for Efficient... | Weihao Cui, Han Zhao, Quan Chen, Hao Wei, Zirui Li, Deze Zeng, Chao Li, Minyi Guo | 18m |
DVABatch is a GPU runtime that slices DNNs into staged executors and exposes three meta-operations new/stress/split to implement a multi-entry multi-exit… | ||
| Serving Heterogeneous Machine Learning Models on Multi-GPU Servers... | Seungbeom Choi, Sunho Lee, Sunho Lee, Yeonjae Kim, Yeonjae Kim, Jongse Park, Youngjin Kwon, Jaehyuk Huh | 17m |
A multi-model scheduling framework implements spatial and temporal packing of GPU resources into configurable virtual partitions called gpulets, profiles… | ||
| PilotFish: Harvesting Free Cycles of Cloud Gaming with Deep Learning Training | Wei Zhang, Binghao Chen, Zhenhua Han, Quan Chen, Peng Cheng, Fan Yang, Ran Shu, Yuqing Yang, Yuqing Yang, Minyi Guo | 19m |
PilotFish co-locates deep learning training with cloud gaming, hooks DirectX/OpenGL to capture idle GPU periods in sub-millisecond granularity, and schedules… | ||
| Sibylla: To Retry or Not To Retry on Deep Learning Job Failure | Taeyoon Kim, Taeyoon Kim, Suyeon Jeong, Suyeon Jung, Jongseop Lee, Soobee Lee, Myeongjae Jeon | 13m |
Sibylla classifies DL job failures as deterministic or nondeterministic by training RNN models (GRU/LSTM-like) on stdout/stderr logs and updating online with… | ||
| Speculative Recovery: Cheap, Highly Available Fault Tolerance with Disaggregated... | Nanqinqin Li, Nanqinqin Li, Anja Kalaba, Michael J. Freedman, Wyatt Lloyd, Amit Levy | 19m |
Speculative recovery claims to cut REDS failover time by running failure detection and recovery in parallel using two new primitives super and collapse… | ||
| Direct Access, High-Performance Memory Disaggregation with DirectCXL | Donghyun Gouk, Sangwon Lee, Sangwon Lee, Miryeong Kwon, Myoungsoo Jung | 12m |
Built a full CXL v2 prototype—DirectCXL prototype—including a CXL root-port CPU, CXL endpoint memory device, reconfigurable switch and a Linux runtime that… | ||
| KSG: Augmenting Kernel Fuzzing with System Call Specification Generation | Hao Sun, Yuheng Shen, Jianzhong Liu, Yiru Xu, Yu Jiang | 20m |
KSG automatically generates Syzlang system call specs by using eBPF/kprobe tracing to extract module entry points, then path-sensitive, symbolic analysis to… | ||
| DLOS: Effective Static Detection of Deadlocks in OS Kernels | Jia-Ju Bai, Tuo Li, Tuo Li, Shi-Min Hu | 20m |
DLOS is a static analysis tool that extracts lock-order constraints via a summary-based lock-usage analysis, detects cycles using a reachability-based… | ||
| Modulo: Finding Convergence Failure Bugs in Distributed Systems with Divergence... | Beom Heyn Kim, Taesoo Kim, David Lie | 19m |
Modulo is a model-based tester that uses Divergence Resync Models to focus state exploration on interleavings of divergence and resync, generating schedules… | ||
| Trustworthy Open Source: The Consequences of Success | Eric Brewer | 49m |
Open source now permeates critical infrastructure with concrete scale: 2.2 trillion downloads last year, 30% single‑maintainer packages and 800,000 new… | ||
| SoftTRR: Protect Page Tables against Rowhammer Attacks using Software-only... | Zhi Zhang, Zhi Zhang, Yueqiang Cheng, Minghua Wang, Minghua Wang, Wei He, Wenhao Wang, Wenhao Wang, Surya Nepal, Yansong Gao, Kang Li, Zhe Wang, Zhe Wang, Chenggang Wu | 18m |
SoftTRR is a software-only defense implemented as a loadable kernel module that protects level-1 page tables by tracking accesses to rows up to six rows away… | ||
| Hardening Hypervisors with Ombro | Ethan Johnson, Colin Pronovost, John Criswell | 23m |
Ombro implements a compiler-targeted virtual ISA (extending SVA/LLVM IR) and ports Xen 4.12 to run under Intel VMX, enforcing CFI plus a first protected shadow… | ||
| HyperEnclave: An Open and Cross-platform Trusted Execution Environment | Yuekai Jia, Shuang Liu, Wenhao Wang, Yu Chen, Zhengde Zhai, Shoumeng Yan, Zhengyu He | 20m |
USENIX ATC '22 - HyperEnclave: An Open and Cross-platform Trusted Execution Environment
Yuekai Jia, Tsinghua University; Shuang Liu, Ant Group; Wenhao Wang… | ||
| PRIDWEN: Universally Hardening SGX Programs via Load-Time Synthesis | Fan Sang, Ming-Wei Shih, Sangho Lee, Xiaokuan Zhang, Michael Steiner, Mona Vij, Taesoo Kim | 19m |
PRIDWEN is a load-time loader that probes remote platform features, decides an optimal set of side‑channel mitigations, and synthesizes a native binary from a… | ||
| Meces: Latency-efficient Rescaling via Prioritized State Migration for Stateful... | Rong Gu, Han Yin, Weichang Zhong, Chunfeng Yuan, Yihua Huang | 17m |
Meces implements fetch-on-demand and prioritized hot-key migration on Apache Flink to rescale stateful stream jobs with minimal latency. It migrates state at… | ||
| DepFast: Orchestrating Code of Quorum Systems | Xuhao Luo, Weihai Shen, Shuai Mu, Tianyin Xu | 21m |
DepFast is a user-space coroutine library that adds a QuorumEvent abstraction to compose RPCs, timeouts, and multiple quorum sizes so code follows the protocol… | ||
| High Throughput Replication with Integrated Membership Management | Pedro Fouto, Nuno Preguiça, João Leitão | 20m |
ChainPaxos is a chain-structured Paxos variant that aggregates accept and commit steps into a pipeline to cut per-node message cost and execute updates with… | ||
| Zero Overhead Monitoring for Cloud-native Infrastructure using RDMA | Zhe Wang, Zhe Wang, Teng Ma, Linghe Kong, Zhenzao Wen, Jingxuan Li, Jingxuan Li, Zhuo Song, Yang Lu, Yong Yang, Yong Yang, Tao Ma, Guihai Chen, Wei Cao | 21m |
Zero collects raw metrics with one-sided RDMA reads into predeclared memory regions so the monitored host shows effectively zero CPU usage, and a single… | ||
| CRISP: Critical Path Analysis of Large-Scale Microservice Architectures | Zhizhou Zhang, Murali Krishna Ramanathan, Prithvi Raj Venkat Raj, Abhishek Parwal, Timothy Sherwood, Milind Chabbi | 19m |
CRISP performs critical path analysis over Jaeger traces across ~40K endpoints at Uber, inferring last-return sinks despite clock skew and missing spans. It… | ||
| Whale: Efficient Giant Model Training over Heterogeneous GPUs | Xianyan Jia, Le Jiang, Ang Wang, Ang Wang, Wencong Xiao, Ziji Shi, Jie Zhang, Xinyuan Li, Xinyuan Li, Langshi Chen, Yong Li, Zhen Zheng, Xiaoyong Liu, Xiaoyong Liu, Wei Lin | 21m |
Whale is a distributed training framework that adds two model-annotation primitives, replicate and split, to transform a local DAG into a distributed plan and… | ||
| IPLFS: Log-Structured File System without Garbage Collection | Juwon Kim, Juwon Kim, Minsu Jang, Muhammad Danish Tehseen, Joontaek Oh, Youjip Won | 19m |
IPLFS implements a log‑structured filesystem that stops doing filesystem‑level garbage collection by setting a very large infinite logical partition (64‑bit… | ||
| Vigil-KV: Hardware-Software Co-Design to Integrate Strong Latency Determinism... | Miryeong Kwon, Seungjun Lee, Seungjun Lee, Hyunkyu Choi, Jooyoung Hwang, Myoungsoo Jung | 20m |
Claims that predictable latency mode (PLM) per NVMe partition and coordinated host scheduling eliminate long-tail reads, implemented by partitioning a 1.92TB… | ||
| Pacman: An Efficient Compaction Approach for Log-Structured Key-Value Store... | Jing Wang, Youyou Lu, Qing Wang, Qing Wang, Minhui Xie, Keji Huang, Jiwu Shu | 20m |
Pacman is a persistent-memory aware compaction design for log-structured key-value stores that offloads reference search to service threads, uses tagged… | ||
| Tetris: Memory-efficient Serverless Inference through Tensor Sharing | Jie Li, Laiping Zhao, Yanan Yang, Yanan Yang, Kunlin Zhan, Keqiu Li | 20m |
Tetris claims an order‑of‑magnitude drop in memory footprint for serverless inference by actually sharing immutable model tensors across co‑located containers… | ||
| PetS: A Unified Framework for Parameter-Efficient Transformers Serving | Zhe Zhou, Xuechao Wei, Jiejing Zhang, Guangyu Sun | 18m |
PetS is a serving framework that decouples PET methods into shared operators and lightweight task-specific PET operators, implements a Python front end with a… | ||
| Campo: Cost-Aware Performance Optimization for Mixed-Precision Neural Network... | Xin He, Jianhua Sun, Hao Chen, Dong Li, Dong Li | 22m |
Campo is a tool that models per-operator FP16 performance and the cost of FP16/FP32 casts using hardware counters, then rewrites TensorFlow graphs to apply… | ||
| Primo: Practical Learning-Augmented Systems with Interpretable Models | Qinghao Hu, Harsha Nori, Peng Sun, Yonggang Wen, Tianwei Zhang, Tianwei Zhang | 19m |
Primo presents two interpretable models PrAM and PrDT, a Distill Engine, and Bayes Optimization to auto-tune pruning and hyperparameters; it adds Monotonic… | ||
| CBMM: Financial Advice for Kernel Memory Managers | Mark Mansi, Bijan Tabatabai, Michael M. Swift | 19m |
CBMM implements a kernel component called the CBMM estimator that consults empirical cost-benefit models and per-workload profiles to decide actions like huge… | ||
| EPK: Scalable and Efficient Memory Protection Keys | Jinyu Gu, Hao Li, Wentai Li, Wentai Li, Yubin Xia, Haibo Chen | 19m |
EPK extends Intel MPK’s 16-domain limit to hundreds by combining MPK with hardware virtualization (EPT + VMFUNC) so each domain is a pair of PK + EPT index… | ||
| Memory Harvesting in Multi-GPU Systems with Hierarchical Unified Virtual Memory | Sangjin Choi, Taeksoo Kim, Taeksoo Kim, Jinwoo Jeong, Rachata Ausavarungnirun, Myeongjae Jeon, Youngjin Kwon, Jeongseob Ahn | 20m |
Implements a hierarchical unified virtual memory (HUVM) that spills GPU pages to neighbor GPUs over NVLink instead of host PCIe, and ships a centralized memory… | ||
| Cachew: Machine Learning Input Data Processing as a Service | Dan Graur, Damien Aymon, Dan Kluser, Tanguy Albrici, Chandramohan A. Thekkath, Ana Klimovic | 19m |
Cachew is a managed service that disaggregates TF Data Service pipelines into dispatcher and stateless workers, adds a caching layer, and applies auto-scaling… | ||
| CoVA: Exploiting Compressed-Domain Analysis to Accelerate Video Analytics | Jinwoo Hwang, Minsu Kim, Minsu Kim, Daeun Kim, Daeun Kim, Seungho Nam, Yoonsung Kim, Yoonsung Kim, Dohee Kim, Dohee Kim, Hardik Sharma, Jongse Park | 18m |
CoVA builds a three-stage pipeline that finds moving-object tracks from motion vectors using a learned model called Plumnet, picks a minimal set of frames via… | ||
| SOTER: Guarding Black-box Inference for General Neural Networks at the Edge | Tianxiang Shen, Ji Qi, Jianyu Jiang, Xian Wang, Xian Wang, Siyuan Wen, Xiyuan Wen, Xusheng Chen, Shixiong Zhao, Sen Wang, Sen Wang, Li Chen, Xiapu Luo, Fengwei Zhang, Heming Cui | 19m |
SOTER is a system implemented in Python with Graphene-SGX that uses parameter-morphed operators to run most linear-heavy layers on untrusted Nvidia GPUs while… | ||
| Hashing Design in Modern Networks: Challenges and Mitigation Techniques | Yunhong Xu, Keqiang He, Rui Wang, Minlan Yu, Nick Duffield, Hassan M. G. Wassel, Shidong Zhang, Shidong Zhang, Leon Poutievski, Junlan Zhou, Amin Vahdat | 20m |
Demonstrates that limited uncorrelated hardware hash functions cause hash polarization that can lose about 33% of usable capacity; builds two deployable fixes… | ||
| Firebolt: Finding Bugs in Programmable Data Plane Generators | Jiamin Cao, Yu Zhou, Chen Sun, Lin He, Zhaowei Xi, Ying Liu | 20m |
Firebolt is a black‑box tester that generates syntax-guided intent generation from a generator’s BNF plus semantic constraints, prunes the intent space with… | ||
| The Computing and Information Science... | Margaret Martonosi | 51m |
CISE manages a $1B+ annual budget, reviews ~7,000 proposals a year and funds research at 373 institutions supporting ~20,000 people and roughly 80% of… | ||
| Investigating Managed Language Runtime Performance: Why JavaScript and Python... | David Lion, Adrian Chiu, Michael Stumm, Ding Yuan | 21m |
Careful measurements compare OpenJDK, Go, V8/Node.js, and CPython against GCC using six reimplemented apps called LangBench and intrusive bytecode-level… | ||
| Automatic Recovery of Fine-grained Compiler Artifacts at the Binary Level | Yufei Du, Ryan Court, Kevin Snow, Fabian Monrose | 21m |
Tool reconstructs per-function compiler passes from binaries using a modified LLVM to generate ground truth, a lightgbm multi-label classifier with seven… | ||
| JITServer: Disaggregated Caching JIT Compiler for the JVM in the Cloud | Alexey Khrabrov, Marius Pirvu, Vijay Sundaresan, Eyal de Lara | 24m |
JIT compilation is moved to a remote service implemented in Eclipse OpenJ9 and augmented with a distributed cache that stores serialized compiled methods with… | ||
| Riker: Always-Correct and Fast Incremental Builds from Simple Specifications | Charlie Curtsinger, Daniel W. Barowy | 23m |
Riker instruments builds with ptrace, seccomp/BPF and an injected library to record system calls into a new TraceIR that models the entire POSIX filesystem… | ||
| FlatFS: Flatten Hierarchical File System Namespace on Non-volatile Memories | Miao Cai, Junru Shen, Bin Tang, Hao Huang, Baoliu Ye | 19m |
FlatFS flattens a POSIX hierarchical namespace into a single indexed namespace and implements a Linux pmfs-based prototype that actually works on Intel Optane… | ||
| StRAID: Stripe-threaded Architecture for Parity-based RAIDs with Ultra-fast SSDs | Shucheng Wang, Qiang Cao, Ziyi Lu, Hong Jiang, Jie Yao, Yuanyuan Dong | 30m |
StRAID implements a one-for-one stripe workers model that replaces MD’s centralized stripe list and heavy spinlock use, adds a parity cache and a two-phase… | ||
| Vinter: Automatic Non-Volatile Memory Crash Consistency Testing for Full Systems | Samuel Kalbfleisch, Lukas Werling, Frank Bellosa | 19m |
Vinter is an automated full-system NVM crash-consistency tester that runs the guest in a VM and records NVM writes, fences and flushes via dynamic binary… | ||
| AlNiCo: SmartNIC-accelerated Contention-aware Request Scheduling for Transaction... | Junru Li, Junru Li, Youyou Lu, Qing Wang, Qing Wang, Jiazhen Lin, Zhe Yang, Zhe Yang, Jiwu Shu | 21m |
USENIX ATC '22 - AlNiCo: SmartNIC-accelerated Contention-aware Request Scheduling for Transaction Processing
Junru Li, Youyou Lu, Qing Wang, Jiazhen Lin, Zhe… | ||
| FpgaNIC: An FPGA-based Versatile 100Gb SmartNIC for GPUs | Zeke Wang, Hongjing Huang, Jie Zhang, fei wu, Gustavo Alonso | 19m |
FpgaNIC is an FPGA SmartNIC that enables GPU virtual address P2P so GPUs talk to NICs over PCIe without CPU copies, implements a reliable 100Gb hardware… | ||
| Faster Software Packet Processing on FPGA NICs with eBPF Program Warping | Marco Bonola, Giacomo Belocchi, Angelo Tulumello, Marco Spaziani Brunella, Giuseppe Siracusano, Giuseppe Bianchi, Roberto Bifulco | 26m |
Program warping statically analyzes unmodified Linux eBPF/XDP bytecode, identifies initial parse-and-key-construction sequences, and replaces them with a… | ||
| NVMe SSD Failures in the Field: the Fail-Stop and the Fail-Slow | Ruiming Lu, Erci Xu, Yiming Zhang, Zhaosheng Zhu, Mengtian Wang, Zongpeng Zhu, Guangtao Xue, Minglu Li, Jiesheng Wu | 20m |
Logs from 1.8 million NVMe SSDs at Alibaba drive these claims: infant mortality disappears (failure rates in months one–three are at or below long‑term rates)… | ||
| CacheSack: Admission Optimization for Google Datacenter Flash Caches | Tzu-Wei Yang, Tzu-Wei Yang, Seth Pollen, Mustafa Uysal, Arif Merchant, Homer Wolfmeister | 16m |
CacheSack is an admission optimizer that partitions traffic by key groups, measures per-category cache benefit with a ghost table, and solves a knapsack… | ||
| Amazon DynamoDB: A Scalable, Predictably Performant, and Fully Managed NoSQL... | Mostafa Elhemali, Niall Gallagher, Somasundaram Perianayagam, Nicholas Gordon, Joseph Idziorek, Richard Krog, Jaso Sorenson, Colin Lazier, Erben Mo, Akhilesh Mritunjai, Somu Perianayagam, Tim Rath, Swami Sivasubramanian, James Christopher Sorenson III, Sroaj Sosothikul, Doug Terry, Akshat Vig | 20m |
Amazon reports DynamoDB serves trillions of API calls and peaks at 89.2 million requests/sec during Prime Day while keeping single‑digit millisecond latencies… | ||
| nothing matches | ||