USENIX Courses

Mirage - A Multi-Level Superoptimizer for Tensor Programs

Discover Mirage, a groundbreaking multi-level superoptimizer for tensor programs using µGraphs to combine algebraic and schedule transformations with custom kernel generation for GPU optimization.

• YouTube
• 16 minutes
• Self-Paced
• Free Video

Turbocharge ANNS on Real Processing-in-Memory by Enabling Fine-Grained Per-PIM-Core Scheduling

Discover how PIMANN revolutionizes Approximate Nearest Neighbor Search using Processing-in-Memory hardware with fine-grained scheduling to achieve 2.4-10.4× throughput improvements.

• YouTube
• 19 minutes
• Self-Paced
• Free Video

Accelerating Nested Virtualization with HyperTurtle

Discover HyperTurtle, a breakthrough approach that accelerates nested virtualization by reducing costly world switches using eBPF programs, achieving up to 27% faster container boot times.

• YouTube
• 16 minutes
• Self-Paced
• Free Video

CrossPipe - Towards Optimal Pipeline Schedules for Cross-Datacenter Training

Discover CrossPipe, a framework optimizing large language model training across geographically distributed datacenters by mitigating network latency and bandwidth limitations.

• YouTube
• 19 minutes
• Self-Paced
• Free Video

MemoryTrap - Booby Trapping Memory to Counter Memory Disclosure Attacks with Hardware Support

Discover MemoryTrap, a hardware-assisted technique using Intel Memory Protection Keys to counter JIT-ROP attacks by placing unreadable "booby traps" in program memory at compile time.

• YouTube
• 15 minutes
• Self-Paced
• Free Video

Low End-to-End Latency atop a Speculative Shared Log with Fix-Ante Ordering

Discover SpecLog, a novel shared log abstraction using fix-ante ordering to reduce end-to-end latency by speculating global order and overlapping computation with coordination.

• YouTube
• 17 minutes
• Self-Paced
• Free Video

EMT - An OS Framework for New Memory Translation Architectures

Explore EMT, a Linux framework enabling diverse memory translation architectures like radix tree and hash table to overcome performance bottlenecks in terabyte-scale memory systems.

• YouTube
• 17 minutes
• Self-Paced
• Free Video

PoWER Never Corrupts - Tool-Agnostic Verification of Crash Consistency and Corruption Detection

Discover novel verification techniques for storage systems resilience using PoWER methodology and corruption detection models in persistent memory environments.

• YouTube
• 18 minutes
• Self-Paced
• Free Video

PMR - Fast Application Response via Parallel Memory Reclaim on Mobile Devices

Discover PMR, a parallel memory reclaim scheme that improves mobile app response times by up to 43.6% through proactive page shrinking and storage-friendly writeback techniques.

• YouTube
• 14 minutes
• Self-Paced
• Free Video

KRR - Efficient and Scalable Kernel Record Replay

Discover KRR, a kernel record-replay framework that efficiently records and reproduces kernel failures with minimal overhead, enabling advanced debugging of complex system bugs.

• YouTube
• 16 minutes
• Self-Paced
• Free Video

Internet Connection Splitting - What's Old is New Again

Explore how performance-enhancing proxies remain relevant for Internet performance despite BBR congestion control and QUIC protocol advances through emulation measurements.

• YouTube
• 14 minutes
• Self-Paced
• Free Video

Inferring Likely Counting-related Atomicity Program Properties for Persistent Memory

Discover how to infer counting-related atomicity properties in persistent memory programs using symbolic range analysis and SMT constraints to detect bugs.

• YouTube
• 17 minutes
• Self-Paced
• Free Video

Swift - Fast Performance Tuning with GAN-Generated Configurations

Discover Swift, a GAN-powered Bayesian optimization approach that accelerates performance tuning for big data systems like Apache Flink and Spark by generating high-quality configurations.

• YouTube
• 18 minutes
• Self-Paced
• Free Video

Converos - Practical Model Checking for Verifying Rust OS Kernel Concurrency

Discover CONVEROS, a practical model-checking methodology for verifying Rust OS kernel concurrency, uncovering 20 critical bugs in ASTERINAS including data races and deadlocks.

• YouTube
• 15 minutes
• Self-Paced
• Free Video

Accelerating Model Training on Ascend Chips - An Industrial System for Profiling, Analysis and Optimization

Discover Hermes, an industrial system for optimizing deep learning model training on Huawei Ascend chips through profiling, bottleneck analysis, and optimization strategies.

• YouTube
• 13 minutes
• Self-Paced
• Free Video