USENIX Courses

RubbleDB - CPU-Efficient Replication with NVMe-oF

Explores CPU-efficient replication in key-value stores using NVMe-oF, introducing novel mechanisms to address challenges and improve performance for write-heavy workloads.

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

AAsclepius - Monitoring, Diagnosing, and Detouring at the Internet Peering Edge

Innovative system for monitoring, diagnosing, and detouring network faults at Internet peering edge, featuring PathDebugging technique to identify fault direction and protect cloud services.

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

Tectonic-Shift - A Composite Storage Fabric for Large-Scale ML Training

Innovative storage fabric design combining flash and HDD tiers to optimize power efficiency and IO performance for large-scale machine learning training, featuring application-aware caching policies.

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

SAGE - Software-based Attestation for GPU Execution

Explore SAGE, a software-based attestation mechanism for secure GPU execution, offering code integrity, computation integrity, and data protection without hardware support.

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

EnvPipe - Performance-Preserving DNN Training Framework for Saving Energy

Innovative framework for energy-efficient DNN training, maximizing power savings while maintaining performance through strategic pipeline scheduling and frequency adjustment.

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

Distributed Transactions at Scale in Amazon DynamoDB

Explore how Amazon DynamoDB implemented distributed transactions without compromising performance, availability, or scale in NoSQL cloud databases using timestamp ordering protocol.

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

UnFaaSener - Latency and Cost Aware Offloading of Functions from Serverless Platforms

Innovative framework for reducing serverless computing costs by offloading functions to non-serverless resources, considering latency, cost, and QoS expectations for optimal decision-making and significant savings.

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

On-demand Container Loading in AWS Lambda

Explore AWS Lambda's innovative storage and caching system for efficient container image delivery, addressing challenges in security, efficiency, latency, and cost at massive scale.

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

Decentralized and Stateful Serverless Computing on the Internet Computer Blockchain

Explore the Internet Computer blockchain's innovative approach to decentralized, stateful serverless computing, addressing scalability, persistence, and deterministic scheduling challenges.

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

oBBR - Optimize Retransmissions of BBR Flows on the Internet

Optimizing BBR algorithm to reduce retransmissions while maintaining high throughput. Adaptive scaling of in-flight data cap and timely bandwidth estimate updates improve Internet streaming performance.

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

EPF - Evil Packet Filter

Explores a new method for bypassing kernel isolation techniques using BPF infrastructure, demonstrating privilege escalation attacks and proposing defenses against such vulnerabilities.

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

Decentralized Application-Level Adaptive Scheduling for Multi-Instance DNNs on Open Mobile Devices

Explores decentralized scheduling for multiple DNN-powered apps on open mobile systems, using Deep Reinforcement Learning to achieve balanced performance gains and energy savings across various scenarios.

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

LoopDelta - Embedding Locality-aware Opportunistic Delta Compression in Inline Deduplication for Highly Efficient Data Reduction

Innovative data compression technique combining deduplication and delta compression for efficient storage. Improves compression ratios and restore performance without affecting backup throughput.

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

Translation Pass-Through for Near-Native Paging Performance in VMs

Innovative memory virtualization mechanism achieving near-native performance in VMs. Enables direct guest-to-host address translation while maintaining isolation through hardware-supported memory tagging.

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

Change Management in Physical Network Lifecycle Automation

Explores automated systems for managing large network lifecycles, focusing on change management challenges and solutions for conflict resolution and model representation updates.

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