USENIX Courses

Efficient Performance-Aware GPU Sharing with Compatibility and Isolation through Kernel Space Interception

Discover KRYPTON, a kernel-space GPU sharing solution that ensures compatibility and isolation while reducing required GPU resources by 32.1% through efficient spatio-temporal sharing.

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

Fast and Synchronous Crash Consistency with Metadata Write-Once File System

Explore a novel file system model achieving fast synchronous crash consistency through metadata write-once approach, optimizing persistent memory I/O performance.

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

NanoFlow - Towards Optimal Large Language Model Serving Throughput

Discover NanoFlow, a novel framework that achieves 1.91× throughput boost for LLM serving by exploiting intra-device parallelism and overlapping heterogeneous operations within GPUs.

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

Disentangling the Dual Role of NIC Receive Rings

Discover how rxBisect decouples memory allocation from packet delivery in NIC receive rings, improving throughput by up to 37% over current methods.

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

Fast Distributed Transactions for RDMA-based Disaggregated Memory

Discover HDTX, a high-performance system that reduces distributed transaction latency by up to 88% through novel RDMA-based designs for disaggregated memory architectures.

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

Cuckoo for Clients - Disaggregated Cuckoo Hashing

Discover RCuckoo, a disaggregated key/value store using cuckoo hashing and one-sided RDMA operations for high-performance data access with enhanced locality and fault tolerance.

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

LITESHIELD - Secure Containers via Lightweight, Composable Userspace μKernel Services

Discover LITESHIELD's innovative userspace microkernel architecture that delivers VM-level security for containers with just 22 syscalls, enabling strong isolation without hypervisor complexity.

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

ZEN - Empowering Distributed Training with Sparsity-driven Data Synchronization

Discover ZEN, a gradient synchronization system that leverages tensor sparsity to accelerate distributed deep learning training by up to 5.09x in communication time.

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

Tintin - A Unified Hardware Performance Profiling Infrastructure to Uncover and Manage Uncertainty

Discover Tintin, a modular hardware performance profiling infrastructure that addresses measurement uncertainty and attribution challenges in microarchitectural event tracking.

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

Picsou - Enabling Replicated State Machines to Communicate Efficiently

Discover Picsou, a breakthrough protocol enabling efficient communication between replicated state machines with 24× performance improvements over existing solutions.

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

MettEagle - Costs and Benefits of Implementing Containers on Microkernels

Explore MettEagle's implementation of containers on microkernels, comparing security benefits and performance against traditional Linux containers in cloud environments.

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

XRT - An Accelerator-Aware Runtime for Accelerated Chip Multiprocessors

Discover XRT, a runtime system designed for accelerated chip multiprocessors that optimizes on-chip accelerator usage to achieve up to 3.2× higher throughput in datacenter applications.

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

QOS - Quantum Operating System

Discover QOS, a modular quantum operating system that manages quantum resources, mitigates hardware errors, and optimizes job scheduling to achieve higher fidelity and utilization.

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

KPerfIR - Towards an Open and Compiler-centric Ecosystem for GPU Kernel Performance Tooling on Modern AI Workloads

Discover KPerfIR, a compiler-centric infrastructure for GPU kernel performance analysis on AI workloads with low overhead and accurate measurements.

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

Neutrino - Fine-grained GPU Kernel Profiling via Programmable Probing

Discover NEUTRINO, a programmable GPU kernel profiler enabling instruction-level analysis across NVIDIA and AMD hardware with minimal overhead for system optimization.

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