USENIX Courses

FLASH - Towards a High-performance Hardware Acceleration Architecture for Cross-silo Federated Learning

Innovative hardware acceleration architecture for cross-silo federated learning, enhancing performance and resource utilization of cryptographic operations while preserving data privacy.

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

Hostping - Diagnosing Intra-host Network Bottlenecks in RDMA Servers

Introduces Hostping, a system for monitoring and diagnosing intra-host network bottlenecks in RDMA servers, addressing performance issues in high-speed networks through innovative loopback testing techniques.

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

Skyplane - Optimizing Transfer Cost and Throughput Using Cloud-Aware Overlays

Optimizes cloud data transfers across regions and providers using network overlays, balancing cost and performance. Outperforms existing services by up to 5x through intelligent path selection and resource allocation.

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

Shockwave - Fair and Efficient Cluster Scheduling for Dynamic Adaptation in Machine Learning

Efficient cluster scheduling system for dynamic adaptation in ML training, optimizing fairness and efficiency through market theory and stochastic dynamic programming.

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

TACCL - Guiding Collective Algorithm Synthesis Using Communication Sketches

Innovative tool for synthesizing efficient communication algorithms in multi-GPU machine learning, outperforming NCCL and accelerating model training across various hardware configurations.

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

ModelKeeper - Accelerating DNN Training via Automated Training Warmup

Accelerate DNN training by repurposing previously-trained models. Identify architectural similarities, select parent models, and perform structure-aware weight transformation for faster training completion without accuracy loss.

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

The Benefit of Hindsight - Tracing Edge-Cases in Distributed Systems

Innovative distributed tracing system for efficiently capturing edge-case issues in large-scale systems, offering low-overhead monitoring and retroactive sampling for precise problem diagnosis.

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

CausalSim - A Causal Framework for Unbiased Trace-Driven Simulation

Innovative framework for unbiased trace-driven simulation, addressing biases in current methods. Learn about CausalSim's causal model approach and its significant improvements in simulation accuracy for various applications.

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

A High-Speed Stateful Packet Processing Approach for Tbps Programmable Switches

Innovative approach for high-speed packet processing using programmable switches, external memory, and processing devices. Enables complex stateful operations at 300G throughput with a single server.

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

Doing More with Less - Orchestrating Serverless Applications without an Orchestrator

Explore application-level orchestration for serverless apps using Unum, offering flexibility, cost-efficiency, and performance gains over standalone orchestrators while maintaining programming interfaces and execution guarantees.

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

μMote - Enabling Passive Chirp De-spreading and μW-level Long-Range Downlink for Backscatter Devices

Innovative long-range receiver for backscatter devices, enabling passive chirp de-spreading and μW-level power consumption. Achieves 400m range at 2kbps with low BER, overcoming previous limitations.

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

Transparent GPU Sharing in Container Clouds for Deep Learning Workloads

Innovative system for efficient GPU sharing in container clouds, enabling higher utilization and faster job completion for deep learning workloads while maintaining performance isolation.

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

RF-Chord - Towards Deployable RFID Localization System for Logistic Networks

Innovative RFID localization system for efficient inventory tracking in logistics. Offers real-time processing, extended range, and reduced errors, enhancing automation in warehouses and delivery stores.

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

Practical Intent-Driven Routing Configuration Synthesis

Introduces Aura, a system for synthesizing datacenter routing policies, featuring RPL language and a compiler for automatic switch configuration generation, supporting frequent policy changes.

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

Enhancing Global Network Monitoring with Magnifier

Innovative approach to enhance global network monitoring by combining sampling with mirroring, improving coverage without increasing overhead. Introduces "negative mirroring" for efficient traffic validation.

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