USENIX Courses

FarReach - Write-back Caching in Programmable Switches

Innovative caching framework for programmable switches, enhancing performance in skewed write-intensive workloads. Addresses challenges of switch programming and data protection while achieving significant throughput gains.

• YouTube
• 19 minutes
• On-Demand
• Free Video

Empowering Azure Storage with RDMA

Exploring Azure's implementation of RDMA for storage, addressing challenges in regional deployment and achieving significant performance improvements and CPU savings.

• YouTube
• 20 minutes
• On-Demand
• Conference Talk

Better Together - Jointly Optimizing ML Collective Scheduling and Execution Planning Using SYNDICATE

Optimizing ML training with SYNDICATE: A framework that minimizes communication bottlenecks and speeds up training for large-scale models through novel motif abstraction and joint optimization techniques.

• YouTube
• 22 minutes
• On-Demand
• Conference Talk

Building Flexible, Low-Cost Wireless Access Networks With Magma

Magma: An open system for building affordable wireless networks in underserved areas, leveraging Internet design patterns and SDN principles to reduce costs and complexity while maintaining key cellular features.

• YouTube
• 20 minutes
• On-Demand
• Conference Talk

Push-Button Reliability Testing for Cloud-Backed Applications with Rainmaker

Explore Rainmaker, a tool for reliability testing of cloud-backed applications. Learn about its fault injection techniques, automatic policies, and success in detecting bugs in popular software.

• YouTube
• 18 minutes
• On-Demand
• Conference Talk

Understanding RDMA Microarchitecture Resources for Performance Isolation

Explores RDMA performance isolation challenges in cloud environments, presenting a model for RNIC resource usage and a test suite to evaluate isolation solutions, offering insights for future designs.

• YouTube
• 14 minutes
• On-Demand
• Conference Talk

Fast, Approximate Vector Queries on Very Large Unstructured Datasets

Explore Auncel, a vector query engine for large datasets, offering bounded errors and latencies. Learn how it outperforms existing solutions in processing high-dimensional data efficiently.

• YouTube
• 14 minutes
• On-Demand
• Conference Talk

Enabling Users to Control Their Internet

CRAB system enables users to control bandwidth allocation across incoming flows without ISP support, improving web page load times and video quality through a novel control loop at receiver devices.

• YouTube
• 17 minutes
• On-Demand
• Conference Talk

Scalable Distributed Massive MIMO Baseband Processing

Innovative system for scaling massive MIMO baseband processing in software-defined radio networks. Reduces inter-server communication to support larger MIMO configurations, enhancing 5G network capacity.

• YouTube
• 17 minutes
• On-Demand
• Conference Talk

BGL - GPU-Efficient GNN Training by Optimizing Graph Data I/O and Preprocessing

Optimizing GPU-based GNN training for large graphs through improved data preprocessing, caching, and partitioning techniques, enhancing efficiency and performance in graph-based machine learning tasks.

• YouTube
• 17 minutes
• On-Demand
• Conference Talk

Synthesizing Runtime Programmable Switch Updates

Explores FlexPlan, a tool for safe runtime updates of programmable switches. Discusses techniques for generating step-by-step plans that maintain correctness and memory constraints during live reconfigurations.

• YouTube
• 15 minutes
• On-Demand
• Conference Talk

Enabling High-Quality Real-Time Communications with Adaptive Frame-Rate

Explore adaptive frame rate control for ultra-low latency in high-quality real-time communications, addressing decoder bottlenecks and network fluctuations to enhance user experience in cloud gaming.

• YouTube
• 15 minutes
• On-Demand
• Conference Talk

SLNet - A Spectrogram Learning Neural Network for Deep Wireless Sensing

Innovative deep learning approach for wireless sensing, combining spectrogram generation and learning to enhance accuracy in various applications like gesture recognition and fall detection.

• YouTube
• 17 minutes
• On-Demand
• Conference Talk

DOTE - Rethinking Predictive WAN Traffic Engineering

Innovative approach to WAN traffic engineering using historical data and deep learning, outperforming existing methods in quality and runtime while nearly matching an omniscient oracle.

• YouTube
• 17 minutes
• On-Demand
• Conference Talk

OpenLoRa - Validating LoRa Implementations through an Extensible and Open-sourced Framework

Explore OpenLoRa, an open-source platform for evaluating LoRa packet collision resolution techniques. Learn about implementation challenges and performance benchmarks in low SNR environments.

• YouTube
• 15 minutes
• On-Demand
• Conference Talk