A Full GPU Virtualization Solution with Mediated Pass-Through
Offered By: USENIX via YouTube
Course Description
Overview
Explore a groundbreaking GPU virtualization solution in this 20-minute conference talk from USENIX ATC '14. Delve into gVirt, a full GPU virtualization implementation that combines native graphics driver support with mediated pass-through technology. Learn how this approach achieves optimal performance, scalability, and secure isolation among guests. Discover the architecture of Intel Processor Graphics and understand key concepts such as pass-through, trap-and-emulation, and render engine sharing. Examine challenges in GPU virtualization, including memory resource partitioning and secure isolation. Analyze performance metrics for both Linux and Windows VMs, and gain insights into scalability and overhead considerations. Gain valuable knowledge about this innovative technology that enables up to 95% native performance for GPU-intensive workloads and efficiently scales up to 7 VMs.
Syllabus
Intro
GPU Usages
Virtualization Usages
Motivations for GPU Virtualization
GPU Virtualization Requirements
GPU Virtualization Approaches
Challenges
Architecture of Intel Processor Graphics
Pass Through
Trap and Emulation
Render Engine Sharing
Display Management
Graphics Memory Resource Partitioning
Address Space Ballooning Inconsistent view due to graphics memory partition
Secure Isolation in gVirt
Vulnerability from Direct Execution
Smart Shadowing
Lazy Shadowing
Configurations
Linux VM Performance
Windows VM Performance
Scalability
Overhead Analysis
Summary
Taught by
USENIX
Related Courses
Amazon DynamoDB - A Scalable, Predictably Performant, and Fully Managed NoSQL Database ServiceUSENIX via YouTube Faasm - Lightweight Isolation for Efficient Stateful Serverless Computing
USENIX via YouTube AC-Key - Adaptive Caching for LSM-based Key-Value Stores
USENIX via YouTube The Future of the Past - Challenges in Archival Storage
USENIX via YouTube A Decentralized Blockchain with High Throughput and Fast Confirmation
USENIX via YouTube