USENIX Courses

Cobra - Making Transactional Key-Value Stores Verifiably Serializable

Explore techniques for verifying serializability in transactional key-value stores, focusing on Cobra's innovative approach to handling real-world workloads and improving verification efficiency.

• YouTube
• 18 minutes
• Self-Paced
• Conference Talk

Fast RDMA-based Ordered Key-Value Store Using Remote Learned Cache

Explores XSTORE, an RDMA-based key-value store using machine learning models as cache for tree-based indexes, improving performance and scalability while efficiently handling dynamic workloads and reducing client-side memory usage.

• YouTube
• 19 minutes
• Self-Paced
• Conference Talk

Semeru - A Memory-Disaggregated Managed Runtime

Innovative distributed JVM for memory-disaggregated environments, featuring a universal Java heap, distributed garbage collection, and efficient swap system to enhance managed cloud application performance.

• YouTube
• 18 minutes
• Self-Paced
• Conference Talk

Thunderbolt - Throughput-Optimized, Quality-of-Service-Aware Power Capping at Scale

Explore power capping in data centers with Thunderbolt, a system optimizing throughput and quality of service. Learn about hardware-agnostic solutions for safe power oversubscription and task-level QoS differentiation.

• YouTube
• 20 minutes
• Self-Paced
• Conference Talk

Twine - A Unified Cluster Management System for Shared Infrastructure

Explore Facebook's Twine, a unified cluster management system for shared infrastructure, scaling to manage one million machines across data centers with innovative approaches to workload customization and resource allocation.

• YouTube
• 19 minutes
• Self-Paced
• Conference Talk

AntMan - Dynamic Scaling on GPU Cluster for Deep Learning

Dynamic scaling system for efficient GPU utilization in deep learning, improving memory and computation usage without compromising fairness or performance in large-scale clusters.

• YouTube
• 18 minutes
• Self-Paced
• Conference Talk

Blockene - A High-throughput Blockchain Over Mobile Devices

Innovative blockchain system enabling smartphone participation in validation and consensus, offering high throughput and scalability while minimizing resource usage through a novel split-trust design.

• YouTube
• 19 minutes
• Self-Paced
• Conference Talk

Calcspar - A Contract-Aware LSM Store for Cloud Storage with Low Latency Spikes

Explores Calcspar, a contract-aware LSM store for cloud storage, addressing latency challenges in Amazon EBS. Demonstrates significant tail latency reduction while maintaining performance in real-world workloads.

• YouTube
• 20 minutes
• Self-Paced
• Conference Talk

Legion - Automatically Pushing the Envelope of Multi-GPU System for Billion-Scale GNN Training

Innovative system for accelerating billion-scale GNN training using multi-GPU caching, hierarchical partitioning, and automatic management, significantly outperforming existing solutions.

• YouTube
• 20 minutes
• Self-Paced
• Conference Talk

AWARE - Automate Workload Autoscaling with Reinforcement Learning in Production Cloud Systems

Explore AWARE, an innovative framework using reinforcement learning for efficient workload autoscaling in cloud systems, offering faster adaptation and improved resource utilization.

• YouTube
• 22 minutes
• Self-Paced
• Conference Talk

VectorVisor - A Binary Translation Scheme for Throughput-Oriented GPU Acceleration

Explore VectorVisor, a novel GPU acceleration approach for server-side applications, offering improved throughput and cost-efficiency through vectorized binary translation and CPU thread emulation.

• YouTube
• 25 minutes
• Self-Paced
• Conference Talk

SecretFlow-SPU - A Performant and User-Friendly Framework for Privacy-Preserving Machine Learning

Explore SecretFlow-SPU, a framework for privacy-preserving machine learning that outperforms existing solutions and simplifies development for users without cryptography expertise.

• YouTube
• 23 minutes
• Self-Paced
• Conference Talk

Beware of Fragmentation - Scheduling GPU-Sharing Workloads with Fragmentation Gradient Descent

Innovative GPU scheduling technique using Fragmentation Gradient Descent to optimize resource allocation, reducing unallocated GPUs by up to 49% and improving utilization in large-scale machine learning clusters.

• YouTube
• 22 minutes
• Self-Paced
• Conference Talk

Sky Computing

Explore the potential evolution of cloud computing ecosystems, drawing parallels with other tech industries and discussing early findings in the field of sky computing.

• YouTube
• 52 minutes
• Self-Paced
• Conference Talk

Bridging the Gap between Relational OLTP and Graph-based OLAP

Explore GART, an innovative system bridging relational OLTP and graph-based OLAP, offering real-time graph analytics with superior performance and data freshness for dynamic applications.

• YouTube
• 20 minutes
• Self-Paced
• Conference Talk