USENIX Courses

CXL-ANNS - Software-Hardware Collaborative Memory Disaggregation and Computation for Billion-Scale Approximate Nearest Neighbor Search

Innovative approach for scalable approximate nearest neighbor search using CXL memory disaggregation, caching, and parallel processing to significantly improve performance and efficiency.

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

Efficient Memory Overcommitment for I/O Passthrough Enabled VMs via Fine-grained Page Meta-data Management

Innovative approach for efficient memory management in virtualized systems, enhancing resource utilization while ensuring DMA safety through fine-grained page meta-data management.

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

MSRL - Distributed Reinforcement Learning with Dataflow Fragments

Explores MSRL, a distributed reinforcement learning system using fragmented dataflow graphs for flexible execution on GPU clusters, offering improved performance and adaptability.

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

LPNS - Scalable and Latency-Predictable Local Storage Virtualization for Unpredictable NVMe SSDs in Clouds

Explore LPNS, a novel system for latency-predictable local storage virtualization in cloud environments, offering improved QoS control and performance for NVMe SSDs.

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

TiDedup - A New Distributed Deduplication Architecture for Ceph

Explore TiDedup, a novel cluster-level deduplication architecture for Ceph, addressing shortcomings in existing designs and improving data reduction, I/O throughput, and reference correction efficiency.

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

Sponge - Fast Reactive Scaling for Stream Processing with Serverless Frameworks

Innovative stream processing system enabling fast reactive scaling for unpredictable workloads using serverless frameworks. Reduces latency and costs while efficiently handling bursty loads in real-time data processing.

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

Accelerating Distributed MoE Training and Inference with Lina

Explore Lina, an innovative system for accelerating distributed Mixture of Experts training and inference, addressing all-to-all communication bottlenecks in large-scale AI models.

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

Bridging the Gap between QoE and QoS in Congestion Control - A Large-scale Mobile Web Service Perspective

Explore a novel approach to enhance mobile web services by bridging Quality of Experience and Quality of Service in congestion control, improving user experience through adaptive algorithms.

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

Confidential Computing within an AI Accelerator

Explore IPU Trusted Extensions for secure AI workloads, offering confidentiality and integrity with low overhead. Learn about hardware isolation, encryption, and multi-party training capabilities in Graphcore's AI accelerators.

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

LLFree - Scalable and Optionally-Persistent Page-Frame Allocation

Innovative page-frame allocator design for improved scalability, memory efficiency, and optional persistence in operating systems. Outperforms Linux's allocator in multi-core systems and heterogeneous memory environments.

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

Light-Dedup - A Light-weight Inline Deduplication Framework for Non-Volatile Memory File Systems

Explore a lightweight deduplication framework for NVM file systems, featuring innovative techniques to enhance performance and reduce costs in next-generation storage media.

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

SingularFS - A Billion-Scale Distributed File System Using a Single Metadata Server

Innovative distributed file system using a single metadata server for billion-scale operations, featuring log-free metadata, hierarchical concurrency control, and hybrid inode partition.

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

Prefix Siphoning - Exploiting LSM-Tree Range Filters for Information Disclosure

Explores key disclosure timing attacks on LSM-tree based key-value stores using range filters, demonstrating security vulnerabilities in data storage systems.

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

Comosum - An Extensible, Reconfigurable, and Fault-Tolerant IoT Platform for Digital Agriculture

Explore an innovative IoT platform for digital agriculture, featuring extensible architecture, fault tolerance, and edge analytics for sustainable farming and increased productivity.

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

STYX - Exploiting SmartNIC Capability to Reduce Datacenter Memory Tax

Innovative framework offloading memory optimization tasks to SmartNICs, reducing server CPU load and cache pollution while maintaining efficiency. Demonstrates significant latency improvements for co-running applications.

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