OSDI 2023
| title | author | length |
|---|---|---|
| Joint Keynote Address | Ion Stoica | 52m |
Claim: a portability layer is infeasible, so Sky runs apps across clouds by matching app requirements to available services via an intercloud broker… | ||
| Ship your Critical Section, Not Your Data: Enabling Transparent Delegation with TCLOCKS | Vishal Gupta, Kumar Kartikeya Dwivedi, Yugesh Kothari, Yueyang Pan, Diyu Zhou, Sanidhya Kashyap | 15m |
TCLocks implements transparent delegation by saving a waiter’s CPU context (IP, SP, registers) at lock and unlocking, making the queue head act as a server… | ||
| RON: One-Way Circular Shortest Routing to Achieve Efficient and Bounded-waiting Spinlocks | Shiwu Lo, Han-Ting Lin, Yao-Hung Hsieh, Chao-Ting Lin, Yu-Hsueh Fang, Ching-Shen Lin, Ching-Chun (Jim) Huang, Kam Yiu Lam, Yuan-Hao Chang | 15m |
RON uses a routing table and precomputes a shortest circular path to hand locks in a one-way circular routing among cores, turning lock handover into an online… | ||
| Userspace Bypass: Accelerating Syscall-intensive Applications | Zhe Zhou, Yanxiang Bi, Junpeng Wan, Yangfan Zhou, Zhou Li | 11m |
Userspace Bypass (UB) transparently moves short user-space instruction sequences into kernel space by identifying hot fast paths, translating them into a… | ||
| Triangulating Python Performance Issues with SCALENE | Emery D. Berger, Sam Stern, Juan Altmayer Pizzorno | 15m |
Scalene is a low‑overhead profiler that simultaneously measures CPU, memory, and GPU usage and distinguishes Python vs native time by using virtual‑time deltas… | ||
| Relational Debugging --- Pinpointing Root Causes of Performance Problems | Xiang (Jenny) Ren, Sitao Wang, Zhuqi Jin, David Lion, Adrian Chiu, Tianyin Xu, Ding Yuan | 15m |
Introduces relations as distributions linking pairs of runtime events and uses causal dependency traversal to pinpoint changes between one good and one bad… | ||
| Accountable authentication with privacy protection: The Larch system for universal login | Emma Dauterman, Danny Lin, Henry Corrigan-Gibbs, David Mazières | 13m |
Larch splits each authentication secret between client and a log server so an attacker who compromises a device cannot authenticate without producing an… | ||
| K9db: Privacy-Compliant Storage For Web Applications By Construction | Kinan Dak Albab, Ishan Sharma, Justus Adam, Benjamin Kilimnik, Aaron Jeyaraj, Raj Paul, Artem Agvanian, Leonhard Spiegelberg, Malte Schwarzkopf | 17m |
K9db is a MySQL‑compatible database that enforces GDPR-style access and deletion by construction: developers add a few schema annotations and K9db builds a… | ||
| Encrypted Databases Made Secure Yet Maintainable | Mingyu Li, Xuyang Zhao, Le Chen, Cheng Tan, Huorong Li, Sheng Wang, Zeyu Mi, Yubin Xia, Feifei Li, Haibo Chen | 14m |
Presents HEDB, a dual‑mode encrypted DB that blocks a newly demonstrated smuggle attack by isolating operators in Execution Mode and enabling safe maintenance… | ||
| LVMT: An Efficient Authenticated Storage for Blockchain | Chenxing Li, Sidi Mohamed Beillahi, Guang Yang, Ming Wu, Wei Xu, Fan Long | 15m |
Claims a storage layer called LVMT that swaps Merkle Patricia Tries for an AMT vector commitment plus a multi-layer trie, stores version numbers instead of… | ||
| Honeycomb: Secure and Efficient GPU Executions via Static Validation | Haohui Mai, Jiacheng Zhao, Hongren Zheng, Yiyang Zhao, Zibin Liu, Mingyu Gao, Cong Wang, Huimin Cui, Xiaobing Feng, Christos Kozyrakis | 14m |
Honeycomb is a software GPU TEE that uses static validation of GPU binaries plus OS-level security monitors to remove the OS and driver from the TCB, prototype… | ||
| An Extensible Orchestration and Protection Framework for Confidential Cloud Computing | Adil Ahmad, Alex Schultz, Byoungyoung Lee, Pedro Fonseca | 15m |
eOPF is a hypervisor-level instrumentation framework that traps enclave events by making enclave pages non-executable, intercepting interrupts and using a… | ||
| Nimble: Rollback Protection for Confidential Cloud Services | Sebastian Angel, Aditya Basu, Weidong Cui, Trent Jaeger, Stella Lau, Srinath Setty, Sudheesh Singanamalla | 14m |
Nimble implements an append-only ledger using a tiny trusted state machine inside TEEs combined with crash-fault-tolerant cloud storage, and it ships a… | ||
| Kerveros: Efficient and Scalable Cloud Admission Control | Sultan Mahmud Sajal, Luke Marshall, Beibin Li, Shandan Zhou, Abhisek Pan, Konstantina Mellou, Deepak Narayanan, Timothy Zhu, David Dion, Thomas Moscibroda, Ishai Menache | 15m |
Kerveros is a Microsoft Azure admission-control system that uses late binding of reservations plus a novel bookkeeping primitive allocable VM to convert… | ||
| Security and Performance in the Delegated User-level Virtualization | Jiahao Chen, Dingji Li, Zeyu Mi, Yuxuan Liu, Binyu Zang, Haibing Guan, Haibo Chen | 12m |
Proposes moving the VM plane out of the kernel so all host‑attacking hypervisor CVEs no longer run in kernel mode, implemented as a 481-line Chisel DV… | ||
| Core slicing: closing the gap between leaky confidential VMs and bare-metal cloud | Ziqiao Zhou, Yizhou Shan, Weidong Cui, Xinyang Ge, Marcus Peinado, Andrew Baumann | 15m |
Proposes partitioning a server into static, hardware-enforced slices that remove the hypervisor from the TCB and eliminate CPU-sharing side channels… | ||
| ExoFlow: A Universal Workflow System for Exactly-Once DAGs | Siyuan Zhuang, Stephanie Wang, Eric Liang, Yi Cheng, Ion Stoica | 16m |
ExoFlow implements a workflow runtime that decouples execution from recovery, providing exactly-once semantics via a separate recovery layer, task annotations… | ||
| Hyrax: Fail-in-Place Server Operation in Cloud Platforms | Jialun Lyu, Marisa You, Celine Irvene, Mark Jung, Tyler Narmore, Jacob Shapiro, Luke Marshall, Savyasachi Samal, Ioannis Manousakis, Lisa Hsu, Preetha Subbarayalu, Ashish Raniwala, Brijesh Warrier, Ricardo Bianchini, Bianca Schroeder, Daniel S. Berger | 15m |
Hyrax lets servers continue hosting VMs with failed but degradable components by disabling faulty parts (using an uncommon rank-bitmask deactivation for DIMMs)… | ||
| NCC: Natural Concurrency Control for Strictly Serializable Datastores by Avoiding the... | Haonan Lu, Shuai Mu, Siddhartha Sen, Wyatt Lloyd | 16m |
NCC is a concretely implemented concurrency control that executes naturally ordered transactions lock-free with non-blocking execution and decoupled response… | ||
| Conveyor: One-Tool-Fits-All Continuous Software Deployment | Boris Grubic, Yang Wang, Tyler Petrochko, Ran Yaniv, Brad Jones, David Callies, Matt Clarke-Lauer, Dan Kelley, Soteris Demetriou, Kenny Yu, Chunqiang Tang | 16m |
Conveyor runs one deployment system across 30,000+ pipelines and ~400,000 deployments per week to millions of hosts, raising fully automated adoption from 12%… | ||
| Chardonnay: Fast and General Datacenter Transactions for On-Disk Databases | Tamer Eldeeb, Xincheng Xie, Philip A. Bernstein, Asaf Cidon, Junfeng Yang | 16m |
Chardonnay is a distributed on-disk key-value store that demonstrates fast datacenter transactions by combining ~150 μs 2PC over Paxos with a tiny low-latency… | ||
| ScaleDB: A Scalable, Asynchronous In-Memory Database | Syed Akbar Mehdi, Deukyeon Hwang, Simon Peter, Lorenzo Alvisi | 16m |
A working in‑memory DB is built that delays range index updates and applies them in batches using per‑thread flushes at merge epochs, storing committed writes… | ||
| VBASE: Unifying Online Vector Similarity Search and Relational Queries via Relaxed... | Qianxi Zhang, Shuotao Xu, Qi Chen, Guoxin Sui, Jiadong Xie, Zhizhen Cai, Yaoqi Chen, Yinxuan He, Yuqing Yang, Fan Yang, Mao Yang, Lidong Zhou | 13m |
VBase identifies a property called relaxed monotonicity that lets approximate nearest-neighbor indices serve as termination checks for SQL-style queries… | ||
| Detecting Transactional Bugs in Database Engines via Graph-Based Oracle Construction | Zu-Ming Jiang, Si Liu, Manuel Rigger, Zhendong Su | 11m |
A tool called TxCheck builds statement-dependency graphs via SQL-level instrumentation and uses transactional oracle construction to generate semantically… | ||
| Take Out the TraChe: Maximizing (Tra)nsactional Ca(che) Hit Rate | Audrey Cheng, David Chu, Terrance Li, Jason Chan, Natacha Crooks, Joseph M. Hellerstein, Ion Stoica, Xiangyao Yu | 15m |
Defines transactional hit rate and builds DeToX, a cache layer that groups keys by dependency (levels and interchangeable groups), scores groups by minimum… | ||
| Replicating Persistent Memory Key-Value Stores with Efficient RDMA Abstraction | Qing Wang, Youyou Lu, Jing Wang, Jiwu Shu | 15m |
OSDI '23 - Replicating Persistent Memory Key-Value Stores with Efficient RDMA Abstraction
Qing Wang, Tsinghua University, Youyou Lu, Tsinghua University, Jing… | ||
| eZNS: An Elastic Zoned Namespace for Commodity ZNS SSDs | Jaehong Min, Chenxingyu Zhao | 14m |
eZNS exposes an adaptive zone abstraction that dynamically reallocates physical active zones across tenants using a centralized zone arbiter and a hierarchical… | ||
| SEPH: Scalable, Efficient, and Predictable Hashing on Persistent Memory | Chao Wang, Junliang Hu, Tsun-Yu Yang, Yuhong Liang, Ming-Chang Yang | 14m |
Implements a new hash table for persistent memory that combines a novel Level Segment (LS) layout, two low‑overhead splits including a one-third split, and… | ||
| No Provisioned Concurrency: Fast RDMA-codesigned Remote Fork for Serverless Computing | Xingda Wei, Fangming Lu, Tianxia Wang, Jinyu Gu, Yuhan Yang, Rong Chen, Haibo Chen | 16m |
MITOSIS implements an OS primitive called remote fork that co-designs Linux kernel behavior with RDMA to let a machine read parent pages on demand and avoid… | ||
| Johnny Cache: the End of DRAM Cache Conflicts (in Tiered Main Memory Systems) | Baptiste Lepers, Willy Zwaenepoel | 16m |
Shows that treating DRAM as a hardware cache works if conflicts are eliminated: two low-overhead techniques—static page-allocation that avoids overlapping… | ||
| TAILCHECK: A Lightweight Heap Overflow Detection Mechanism with Page Protection and... | Amogha Udupa Shankaranarayana Gopal, Raveendra Soori, Michael Ferdman, Dongyoon Lee | 15m |
Implements a runtime that pairs each allocation with a page-aligned shadow TailObject and stores the distance to that shadow in the pointer’s high bits as a… | ||
| SMART: A High-Performance Adaptive Radix Tree for Disaggregated Memory | Xuchuan Luo, Pengfei Zuo, Jiacheng Shen, Jiazhen Gu, Xin Wang, Michael R. Lyu, Yangfan Zhou | 14m |
Builds a radix tree tuned for disaggregated memory and actually implements it: internal pointers live in fixed-size slots updated atomically via RDMA… | ||
| ORC: Increasing Cloud Memory Density via Object Reuse with Capabilities | Vasily A. Sartakov, Lluís Vilanova, Munir Geden, David Eyers, Takahiro Shinagawa, Peter Pietzuch | 15m |
ORC builds a runtime that reuses immutable binary objects across tenants by using hardware memory capabilities (ARM/CHERI-style) and a new relocation type to… | ||
| Global Capacity Management With Flux | Marius Eriksen, Kaushik Veeraraghavan, Yusuf Abdulghani, Andrew Birchall, Po-Yen Chou, Richard Cornew, Adela Kabiljo, Ranjith Kumar S, Maroo Lieuw, Justin J. Meza, Scott Michelson, Thomas Rohloff, Hayley Russell, Jeff Qin, Chunqiang Tang | 16m |
Flux automates regional capacity allocation using RPC tracing to build per-service capacity models, then runs a joint capacity and traffic solver that… | ||
| Defcon: Preventing Overload with Graceful Feature Degradation | Justin J. Meza, Thote Gowda, Ahmed Eid, Tomiwa Ijaware, Dmitry Chernyshev, Yi Yu, Md Nazim Uddin, Rohan Das, Chad Nachiappan, Sari Tran, Shuyang Shi, Tina Luo, David Ke Hong, Sankaralingam Panneerselvam, Hans Ragas, Svetlin Manavski, Weidong Wang, Francois Richard | 15m |
Defcon is a system that lets operators disable noncritical features to prevent overload by trading user impact for capacity savings, using Defcon knob actuator… | ||
| Cilantro: Performance-Aware Resource Allocation for General Objectives via Online... | Romil Bhardwaj, Kirthevasan Kandasamy, Asim Biswal, Wenshuo Guo, Ben Hindman, Joseph E. Gonzalez, Michael Jordan, Ion Stoica | 16m |
Cilantro is a scheduler that replaces user resource requests with online learning from real-time job feedback to map resources to performance, decouples… | ||
| Karma: Resource Allocation for Dynamic Demands | Midhul Vuppalapati, Giannis Fikioris, Rachit Agarwal, Asaf Cidon, Anurag Khandelwal, Eva Tardos | 17m |
Karma introduces credits that track donated and borrowed slices and prioritizes borrowers with the most credits to balance allocations over time, proving… | ||
| AlpaServe: Statistical Multiplexing with Model Parallelism for Deep Learning Serving | Zhuohan Li, Lianmin Zheng, Yinmin Zhong, Vincent Liu, Ying Sheng, Xin Jin, Yanping Huang, Zhifeng Chen, Hao Zhang, Joseph E. Gonzalez, Ion Stoica | 15m |
AlpaServe uses model parallelism not just to fit big models but to statistically multiplex GPUs across many models, trading partition overhead for burst… | ||
| Cocktailer: Analyzing and Optimizing Dynamic Control Flow in Deep Learning | Chen Zhang, Lingxiao Ma, Jilong Xue, Yining Shi, Ziming Miao, Fan Yang, Jidong Zhai, Zhi Yang, Mao Yang | 14m |
OSDI '23 - Cocktailer: Analyzing and Optimizing Dynamic Control Flow in Deep Learning
Chen Zhang, Tsinghua University, Lingxiao Ma, Microsoft Research, Jilong… | ||
| Welder: Scheduling Deep Learning Memory Access via Tile-graph | Yining Shi, Zhi Yang, Jilong Xue, Lingxiao Ma, Yuqing Xia, Ziming Miao, Yuxiao Guo, Fan Yang, Lidong Zhou | 14m |
Welder is a deep-learning compiler that builds a tile-graph abstraction to connect operators at register/shared-memory levels and schedule tiling across… | ||
| Effectively Scheduling Computational Graphs of Deep Neural Networks toward Their... | Jie Zhao, Siyuan Feng, Xiaoqiang Dan, Fei Liu, Chengke Wang, Sheng Yuan, Wenyuan Lv, Qikai Xie | 15m |
A new scheduler partitions DNN computation graphs into coarse-grained subgraphs that fuse operators into larger kernels to keep tensors in local buffers and… | ||
| EINNET: Optimizing Tensor Programs with Derivation-Based Transformations | Liyan Zheng, Haojie Wang, Jidong Zhai, Muyan Hu, Zixuan Ma, Tuowei Wang, Shuhong Huang, Xupeng Miao, Shizhi Tang, Kezhao Huang, Zhihao Jia | 15m |
EinNet claims a larger optimizer search space by turning tensor programs into tensor expressions and applying derivation-based tensor expression… | ||
| Hydro: Surrogate-Based Hyperparameter Tuning Service in Datacenters | Qinghao Hu, Zhisheng Ye, Meng Zhang, Qiaoling Chen, Peng Sun, Yonggang Wen, Tianwei Zhang | 16m |
Hydro builds a surrogate-based tuning service that trains scaled-down models via Torch FX, applies mu parameterization automatically, fuses multiple trials… | ||
| MGG: Accelerating Graph Neural Networks with Fine-Grained Intra-Kernel Communication... | Yuke Wang, Boyuan Feng, Zheng Wang, Tong Geng, Kevin Barker, Ang Li, Yufei Ding | 15m |
MGG is a full-graph GNN runtime that implements a fine-grained dynamic software pipeline inside GPU kernels to overlap remote-label fetch, local neighbor… | ||
| Optimizing Dynamic Neural Networks with Brainstorm | Weihao Cui, Zhenhua Han, Lingji Ouyang, Yichuan Wang, Ningxin Zheng, Lingxiao Ma, Yuqing Yang, Fan Yang, Jilong Xue, Lili Qiu, Lidong Zhou, Quan Chen, Haisheng Tan, Minyi Guo | 13m |
Brainstorm introduces Cell as the sub-tensor routing unit and Router as a programmable dispatch operator, collects fine-grained runtime profiles, and applies… | ||
| AdaEmbed: Adaptive Embedding for Large-Scale Recommendation Models | Fan Lai, Wei Zhang, Liu Rui, William Tsai, Xiaohan Wei, Yuxi Hu, Sabin Devkota, Jianyu Huang, Jongsoo Park, Xing Liu, Zeliang Chen, Ellie Wen, Paul Rivera, Jie You, Chun-cheng Jason Chen, Mosharaf Chowdhury | 15m |
AdaEmbed is a system that performs in-training embedding pruning for DLRMs, ranking billions of embedding rows by runtime access frequency and a moving average… | ||
| BWoS: Formally Verified Block-based Work Stealing for Parallel Processing | Jiawei Wang, Bohdan Trach, Ming Fu, Diogo Behrens, Jonathan Schwender, Yutao Liu, Jitang Lei, Viktor Vafeiadis, Hermann Härtig, Haibo Chen | 15m |
Presents Block-based Work Stealing (BWoS), a queue design that splits per-core queues into fixed-size blocks so owners work inside a block using relaxed… | ||
| Spoq: Scaling Machine-Checkable Systems Verification in Coq | Xupeng Li, Xuheng Li, Wei Qiang, Ronghui Gu, Jason Nieh | 13m |
Spoq produces machine-checkable Coq proofs by compiling unmodified C to LLVM IR and then translating LLVM into Coq with full C semantics so macros, inline… | ||
| Verifying vMVCC, a high-performance transaction library using multi-version concurrency.. | Yun-Sheng Chang, Ralf Jung, Upamanyu Sharma, Joseph Tassarotti, M. Frans Kaashoek, Nickolai Zeldovich | 14m |
vMVCC claims to be the first MVCC transaction library with a machine-checked proof and ships as about 800 lines of Go (open source). It implements in-memory… | ||
| Automated Verification of Idempotence for Stateful Serverless Applications | Haoran Ding, Zhaoguo Wang, Zhuohao Shen, Rong Chen, Haibo Chen | 14m |
Flux presents a verifier and advisor that check and fix idempotence for stateful serverless apps by proving a new correctness notion idempotence consistency… | ||
| Sharding the State Machine: Automated Modular Reasoning for Complex Concurrent System | Travis Hance, Yi Zhou, Andrea Lattuada, Reto Achermann, Alex Conway, Ryan Stutsman, Gerd Zellweger, Chris Hawblitzel, Jon Howell, Bryan Parno | 13m |
IronSync is an automated verification framework that uses an ownership type system and a Localized Transition System to split global proofs into small… | ||
| Flor: An Open High Performance RDMA Framework Over Heterogeneous RNICs | Qiang Li, Yixiao Gao, Xiaoliang Wang, Haonan Qiu, Yanfang Le, Derui Liu, Qiao Xiang, Fei Feng, Peng Zhang, Bo Li, Jianbo Dong, Lingbo Tang, Hongqiang Harry Liu, Shaozong Liu, Weijie Li, Rui Miao, Yaohui Wu, Zhiwu Wu, Chao Han, Lei Yan, Zheng Cao, Zhongjie Wu, Chen Tian, Guihai Chen, Dennis Cai, Jinbo Wu, Jiaji Zhu | 15m |
Flor is an RDMA framework that keeps the RoCEv2 data path in RNIC hardware but moves vendor-specific control logic into software running on NPUs/DPUs/host… | ||
| ShRing: Networking with Shared Receive Rings | Boris Pismenny, Adam Morrison, Dan Tsafrir | 14m |
Multicore NIC receive rings default to 1024-entry rings with ~1.5 KiB per buffer, so per-core working sets reach ~1.5 MiB and the aggregate exceeds LLC/DDIO… | ||
| ServiceRouter: Hyperscale and Minimal Cost Service Mesh | Harshit Saokar, Soteris Demetriou, Nick Magerko, Max Kontorovich, Josh Kirstein, Margot Leibold, Dimitrios Skarlatos, Hitesh Khandelwal, Chunqiang Tang | 17m |
ServiceRouter runs Meta’s global service mesh with millions of L7 routers and routes tens of billions of RPCs per second while keeping the main data plane… | ||
| Characterizing Off-path SmartNIC for Accelerating Distributed Systems | Xingda Wei, Rongxin Cheng, Yuhan Yang, Rong Chen, Haibo Chen | 15m |
A first holistic measurement of the off‑path SmartNIC BlueField-2 maps actual communication paths between client, on‑board SoC, NIC and host and finds concrete… | ||
| Ensō: A Streaming Interface for NIC-Application Communication | Hugo Sadok, Nirav Atre, Zhipeng Zhao, Daniel S. Berger, James C. Hoe, Aurojit Panda, Justine Sherry, Ren Wang | 15m |
Replaces packetized NIC-CPU buffers with a contiguous ring streaming pipe that presents an unbounded byte stream, removing per-packet descriptors and enabling… | ||
| nothing matches | ||