USENIX Courses

Habitat - A Runtime-Based Computational Performance Predictor for Deep Neural Network Training

Explore a runtime-based technique for predicting GPU performance in deep neural network training, aiding cost-efficient hardware selection for researchers and practitioners.

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

ZeRO-Offload - Democratizing Billion-Scale Model Training

ZeRO-Offload: Enabling billion-scale model training on a single GPU by offloading data and compute to CPU, minimizing data movement, and optimizing efficiency. Scales to multiple GPUs and works with model parallelism.

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

SKQ - Event Scheduling for Optimizing Tail Latency in a Traditional OS Kernel

Explore SKQ, a FreeBSD Kqueue redesign that significantly improves application tail latency and low-latency throughput through scalable architecture and event scheduling policies.

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

Improving Performance of Flash Based Key-Value Stores Using Storage Class Memory as a Volatile Memory Extension

Explore using Storage Class Memory to enhance flash-based key-value stores' performance in data centers, focusing on RocksDB improvements and cost-effectiveness for large-scale applications.

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

A Linux Kernel Implementation of the Homa Transport Protocol

Explore Homa/Linux: a high-performance transport protocol outperforming TCP and DCTCP in latency and congestion control, with potential for further optimization through software overhead reduction.

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

MapperX - Adaptive Metadata Maintenance for Fast Crash Recovery of DM-Cache Based Hybrid Storage Devices

Explores MapperX, an extension to DM-cache that uses adaptive bit-trees for faster crash recovery in hybrid storage devices, balancing performance and data integrity.

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

Fair Scheduling for AVX2 and AVX-512 Workloads

Explore fair CPU scheduling for AVX2 and AVX-512 workloads, addressing power-limited CPUs and frequency reduction impacts. Learn techniques to improve performance isolation and fairness in multi-task environments.

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

SONIC - Application-Aware Data Passing for Chained Serverless Applications

Optimizes data passing in serverless workflows, adapting to dynamic factors for improved performance and cost-efficiency in analytics applications.

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

Exploring the Design Space of Page Management for Multi-Tiered Memory Systems

Explores page management strategies for multi-tiered memory systems, analyzing access locality and tier importance to optimize performance in DRAM and SCM configurations.

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

UniStore - A Fault-Tolerant Marriage of Causal and Strong Consistency

Explores a fault-tolerant data store combining causal and strong consistency for geo-distributed systems, addressing liveness challenges and optimizing durability costs in multi-consistency environments.

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

ZNS - Avoiding the Block Interface Tax for Flash-based SSDs

Explore Zoned Namespace SSDs: improved efficiency, higher throughput, and lower latency for flash storage. Learn about interface changes, software adaptations, and performance benefits in Linux and RocksDB.

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

Faastlane - Accelerating Function-as-a-Service Workflows

Accelerating FaaS workflows by executing functions as threads, providing lightweight isolation, and dynamically identifying parallelism opportunities to enhance performance and reduce latency.

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

ArchTM - Architecture-Aware, High Performance Transaction for Persistent Memory

Innovative architecture-aware transaction system for persistent memory, enhancing performance through strategic write management and memory allocation techniques.

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

The Good, the Bad and the Ugly - The 3 Learnings of an SRE

Explore core SRE principles, focusing on incident management, monitoring, and proactive approaches. Gain insights from a young engineer's journey transitioning from academia to industry.

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

Building Actionable Code Ownership

Discover Pinterest's scalable API ownership framework, designed to manage 1700+ endpoints across 70+ teams, addressing challenges and implementing effective solutions for large-scale code management.

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