OSDI (Operating Systems Design and Implementation) Courses

USENIX

OSDI '22 - Immortal Threads: Multithreaded Event-driven Intermittent Computing on Ultra-Low-Power...

Syllabus: Intro Energy Harvesting Batteryless Devices Power Failures - Intermittent Execution Checkpoints vs Tasks Event Handling Complexity Limited Concurrency Compiler Frontend Enabling Micro Continuations Almost Free Checkpoints Thread Scheduling Conc…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - SparTA: Deep-Learning Model Sparsity via Tensor-with-Sparsity-Attribute

Syllabus: Intro Computation Capacity vs DNN Model Size Sparsity Commonly Exists Evolving of Sparsity Pattern Obstacles of Sparsity Optimization The Myth of Proxy Metrics Across-Stack Innovations in Silos SparTA: An End-to-End Approach to Model Sparsity C…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - Carbink: Fault-Tolerant Far Memory

Syllabus: Intro Memory-Intensive Applications in Data Centers Application Interface: Remotable Pointers Replication vs. Erasure Coding Challenge 1: Remotable Objects Have Different Sizes Challenge 2: Efficient Swapping under Erasure Codir Remote Compacti…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - CAP-VMs: Capability-Based Isolation and Sharing in the Cloud

Syllabus: Intro Imperial College London Clouds: Isolation vs. Sharing VMs: Strong, Heavyweight Isolation Containers: Weak, Lightweight Isolation VMs & Containers: The MMU Tax CHERI Capabilities Challenges for Cloud Stacks with Hardware Capabilities CVM:…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - Orca: A Distributed Serving System for Transformer-Based Generative Models

Syllabus: Intro Generative Models Inference of Generative Language M Serving of Generative Language Mo Problem 1: Request-Level Schedulin Solution 1: Iteration-Level Schedulin Problem 2: Batching Solution 2: Selective Batching Orca System Architecture Sc…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - MemLiner: Lining up Tracing and Application for a Far-Memory-Friendly Runtime

Syllabus: Intro Memory Capacity Bottleneck in Datace Far-Memory System High-level Languages Garbage Collection Resource Competition Ineffective Prefetching Can we disable concurrent tracing? Observations Key Design Idea Object Classification Challenges i…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - Owl: Scale and Flexibility in Distribution of Hot Content

Syllabus: Intro Content distribution: a recurring A lot of data to distribute Exacting requirements Prior approaches Hierarchical caching Owl: Centralization vs. decentrali Owl: Ephemeral distribution tree Scaling the control plane Tracker sharding examp…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - Unity: Accelerating DNN Training Through Joint Optimization of Algebraic...

Syllabus: Introduction Unitys Goal Parallelization Parallel Computation Graph Data Parallelization PCG Advantages Techniques Results Conclusion

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - Metastable Failures in the Wild

Syllabus: Intro What are Metastable Failures? Metastable Failures are Prevalent Metastability in the Wild - Survey Defining Metastability - System States Survey Summary Metastability Taxonomy - Trigger Metastability Taxonomy - Sustaining ef Four Metastab…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - KSplit: Automating Device Driver Isolation

Syllabus: Intro Driver vulnerabilities Driver isolation architecture Isolation performance Challenge: Large interface boundary Challenge: Complex data structures Challenge: Low-level kernel/C idioms Challenge: Concurrency primitives KSplit design choices…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - ROLLER: Fast and Efficient Tensor Compilation for Deep Learning

Syllabus: Intro Excessive Compilation Time Black-Box Compiler Motivating Example: 8k^3 matmul Roller: A White-Box Solution Improving Pipeline Throughput Abstracted GPU (V100 Example) Small & Irregular Shapes Evaluations - V100 Performance Evaluation Comp…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '22 - Verifying Hardware Security Modules with Information-Preserving Refinement

Syllabus: Intro HSMs: powerful tools for securing s HSMs suffer from bugs Goal: HSMs without security vulnera Approach: formal verification Example: PIN-protected backup HS How to relate implementation to spe Information-preserving refinement IPR: driver…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '20 - Serving DNNs like Clockwork: Performance Predictability from the Bottom Up

Syllabus: Introduction High Tail Latencies Predictable Worker Clockwork Clockwork Example Conclusion

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '20 - Efficiently Mitigating Transient Execution Attacks using the Unmapped Speculation...

Syllabus: Intro Transient execution attacks risk leaking information Linux maintains security using software mitigations Software mitigations are expensive Goal: faster mitigations Transient execution attack example Typical mitigation approach Ward has a…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course

USENIX

OSDI '20 - Heterogeneity-Aware Cluster Scheduling Policies for Deep Learning Workloads

Syllabus: Intro Hardware for ML training is becoming highly specialized and heterogeneous! How should we allocate heterogeneous resources? Challenge 1: Heterogeneous performance Challenge 2: Diverse scheduling objectives Related work Gavel: A new heterog…

• YouTube
• Less than 1 hour of material
• On-Demand
• Free Online Course