USENIX Courses

Learning Production-Optimized Congestion Control Selection for Alibaba Cloud CDN

Discover how AliCCS revolutionizes CDN performance by intelligently selecting optimal congestion control algorithms, achieving 9.31% QoE improvement and massive cost savings.

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

Holmes - Localizing Irregularities in LLM Training with Mega-scale GPU Clusters

Discover Holmes, a system that localizes training irregularities in large-scale GPU clusters for LLMs, achieving 97.21% accuracy and 6.52× speedup over traditional methods.

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

Securing Public Cloud Networks with Efficient Role-based Micro-Segmentation

Discover how to implement efficient role-based micro-segmentation for securing public cloud networks at scale using novel network-flow analysis techniques.

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

Everything Matters in Programmable Packet Scheduling

Discover PACKS, a novel approximate PIFO scheduler that improves programmable packet scheduling by addressing both scheduling behavior and admission control on existing hardware.

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

SIRD - A Sender-Informed, Receiver-Driven Datacenter Transport Protocol

Discover SIRD, a novel datacenter transport protocol that combines receiver-driven scheduling with sender feedback to maximize bandwidth utilization while minimizing latency and buffering.

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

Mowgli - Passively Learned Rate Control for Real-Time Video

Discover Mowgli, a data-driven rate control system that improves video conferencing quality by learning from existing telemetry logs without performance degradation during training.

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

Pyrrha - Congestion-Root-Based Flow Control to Eliminate Head-of-Line Blocking in Datacenter

Discover Pyrrha, a scalable flow control approach that eliminates head-of-line blocking in datacenters using congestion-root-based methods with minimal queues.

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

Efficient Multi-WAN Transport for 5G with OTTER

Discover OTTER's multi-WAN optimization for 5G networks, achieving 13% more throughput, 15% less RTT, and 45% less jitter across cloud-edge infrastructures.

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

Quicksand - Harnessing Stranded Datacenter Resources with Granular Computing

Discover Quicksand, a framework that eliminates datacenter resource waste through granular computing, splitting applications into resource proclets for millisecond migration and improved utilization.

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

High-Level Programming for Application Networks

Discover AppNet, a framework for building expressive, high-performance application networks with high-level programming that reduces RPC latency by up to 82% compared to traditional service meshes.

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

PAPAYA Federated Analytics Stack - Engineering Privacy, Scalability and Practicality

Discover Meta's PAPAYA system for federated analytics, combining trusted execution environments with on-device computing to enable privacy-preserving data insights at scale.

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

Enhancing Network Failure Mitigation with Performance-Aware Ranking

Discover advanced network failure mitigation techniques using performance-aware ranking to optimize end-to-end flow metrics in cloud datacenters for improved throughput.

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

Evolution of Aegis - Fault Diagnosis for AI Model Training Service in Production

Discover how Alibaba Cloud evolved Aegis, a fault diagnosis system for AI model training that reduces idle time by 97% and task restarts by 84% in production environments.

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

AsTree - An Audio Subscription Architecture Enabling Massive-Scale Multi-Party Conferencing

Discover AsTree, ByteDance's innovative audio subscription architecture that dramatically reduces signaling overhead and bandwidth consumption in massive-scale video conferencing systems.

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

Enabling Portable and High-Performance SmartNIC Programs with Alkali

Discover how Alkali framework enables portable SmartNIC programming across different hardware with automatic optimization, achieving near-expert performance levels.

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