USENIX Courses

GPHash: An Efficient Hash Index for GPU with Byte-Granularity Persistent Memory

Discover GPHash, an efficient hash index for GPU with persistent memory that uses lock-free execution, low-overhead consistency guarantees, and smart caching to outperform existing approaches by up to 27.62× on real-world workloads.

• YouTube
• 14 minutes
• On-Demand
• Free Video

HaSiS: A Hardware-assisted Single-index Store for Hybrid Transactional and Analytical Processing

Explore HaSiS, a novel hardware-assisted single-index store design for HTAP workloads that leverages computational storage drives with built-in compression to overcome traditional multi-index limitations and deliver instant data freshness.

• YouTube
• 15 minutes
• On-Demand
• Free Video

Oasis: An Out-of-core Approximate Graph System via All-Distances Sketches

Explore Oasis, an Out-of-core Approximate Graph System that leverages All-Distances Sketches for efficient large-scale graph processing with minimal memory requirements while maintaining high accuracy.

• YouTube
• 19 minutes
• On-Demand
• Free Video

Don't Maintain Twice, It's Alright: Merged Metadata Management in Deduplication File System with GogetaFS

Explore GogetaFS, a deduplication file system that merges metadata management to reduce overhead by up to 38%, achieving better performance through logical-fingerprint-physical mapping.

• YouTube
• 18 minutes
• On-Demand
• Free Video

Directory-Granularity Filesystem Journaling for CMM-H SSDs

Explore DJFS, a journaling filesystem with per-directory transactions that improves throughput by reducing lock contention, transaction conflicts, and enabling parallel journal commits for better performance on SSDs.

• YouTube
• 16 minutes
• On-Demand
• Free Video

Silhouette: Leveraging Consistency Mechanisms to Detect Bugs in Persistent Memory-Based File Systems

Explore Silhouette, a framework detecting bugs in persistent memory file systems by leveraging consistency mechanisms to reduce exploration space and find bugs faster than existing tools.

• YouTube
• 16 minutes
• On-Demand
• Free Video

Archer: Adaptive Memory Compression with Page-Association-Rule Awareness for High-Speed Response of Mobile Devices

Discover Archer, an innovative memory compression framework for mobile systems that uses flexible granularity and association-rule mining to significantly improve app launching speed, photographic speed, and frame rate.

• YouTube
• 16 minutes
• On-Demand
• Free Video

LeapGNN: Accelerating Distributed GNN Training Leveraging Feature-Centric Model Migration

Discover LeapGNN, a feature-centric framework that accelerates distributed GNN training by bringing models to vertex features, reducing communication bottlenecks through micrograph-based training and feature pre-gathering techniques.

• YouTube
• 14 minutes
• On-Demand
• Free Video

IMPRESS: An Importance-Informed Multi-Tier Prefix KV Storage System for Large Language Model Inference

Discover IMPRESS, a multi-tier prefix KV storage system that reduces time to first token in LLM inference by identifying and loading only important prefix KVs, achieving up to 2.8× faster performance while maintaining accuracy.

• YouTube
• 13 minutes
• On-Demand
• Free Video

AegonKV: A High Bandwidth, Low Tail Latency, and Low Storage Cost KV-Separated LSM Store

Discover AegonKV, a SmartSSD-based key-value store that significantly improves throughput, reduces tail latency, and minimizes storage costs through innovative garbage collection offloading for LSM-based systems.

• YouTube
• 16 minutes
• On-Demand
• Free Video

HiDPU: A DPU-Oriented Hybrid Indexing Scheme for Disaggregated Storage Systems

Explore HiDPU, a hybrid indexing scheme for DPUs in disaggregated storage that reduces memory usage by 92% and improves query performance up to 6.3x through multi-level indexing and optimized DPU-host interactions.

• YouTube
• 18 minutes
• On-Demand
• Free Video

Cloudscape: A Study of Storage Services in Modern Cloud Architectures

Explore Cloudscape, a comprehensive analysis of storage services in AWS cloud architectures, revealing usage patterns, service prevalence, and integration with computing services like Lambda and EC2.

• YouTube
• 17 minutes
• On-Demand
• Free Video

Maat: Analyzing and Optimizing Overcharge on Blockchain Storage

Explore Maat, a tool addressing blockchain storage overcharging issues through fine-grained data collection, consensus-oriented optimizations, and resource pre-allocation, reducing Ethereum transaction fees by 32%.

• YouTube
• 14 minutes
• On-Demand
• Free Video

Towards High-throughput and Low-latency Billion-scale Vector Search via CPU/GPU Collaborative Filtering and Re-ranking

Explore FusionANNS, a novel system for billion-scale vector search that achieves high throughput, low latency, and cost efficiency through CPU/GPU collaborative filtering and innovative I/O optimization techniques.

• YouTube
• 16 minutes
• On-Demand
• Free Video

Mooncake: Trading More Storage for Less Computation — A KVCache-centric Architecture for Serving LLM Chatbot

Explore Mooncake, a KVCache-centric architecture for LLM chatbots that increases request capacity by 59-498% through efficient resource utilization and innovative scheduling while maintaining strict latency SLOs.

• YouTube
• 17 minutes
• On-Demand
• Free Video