GPU Virtualization Shadow GTT Scalability

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current GPU virtualization solutions, such as gVirt, face limitations in scalability, allowing only up to 3 guest vGPU instances per physical Intel GPU, leading to severe GPU resource underutilization and inefficiencies in consolidating VM instances with varying workload patterns.

Innovation Solution

The gScale system introduces a dynamic sharing scheme using private shadow graphics translation tables (GTTs) and predictive GTT copying to enable multiple vGPUs to share an overlapped range of global graphics memory space, allowing for a higher density of vGPU instances and optimizing resource utilization through ladder mapping and fence memory space pooling.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If gVirt is used to provide native graphics driver support in each VM, then performance is optimized, but scalability is limited to only 3 vGPU instances per physical GPU

Engineering Contradiction:

ImproveperformanceVSAvoidscalability

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent segments the graphics memory space into multiple virtual address spaces, each mapped to a separate vGPU instance. By dividing the physical GPU resources and creating independent virtual address space mappings, the system can support more than 3 vGPU instances while maintaining performance optimization through native driver support in each VM.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If the number of vGPU instances is increased to improve scalability, then resource consolidation improves, but GPU resource underutilization occurs with current gVirt limitations

Engineering Contradiction:

ImprovescalabilityVSAvoidGPU resource underutilization

Core Design Contradiction:

Adaptability or versatilityVSLoss of energy

Solution Approach 1:

The patent creates a universal virtualization layer that can accommodate multiple vGPU instances with different workload patterns on a single physical GPU. The multi-functionality is achieved through dynamic resource allocation and context switching mechanisms that allow the same physical GPU to serve multiple virtual instances efficiently, improving both scalability and utilization.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Adaptability or versatility

If more guest VMs are consolidated to a single host, then resource multiplexing opportunities increase, but current gVirt scalability limits the number of VM instances

Engineering Contradiction:

Improveresource consolidationVSAvoidVM instance density

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary virtualization layer that mediates between multiple guest VMs and the physical GPU. This intermediary layer handles context switching, resource allocation, and address space management, enabling high density of VM instances without directly increasing the complexity of individual VM-GPU interactions. The intermediary abstracts the complexity while enabling scalable consolidation.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS10430991B2System, apparatus, and method for optimizing a scalable GPU virtualization

Publication Date: 2019.10.01 SHANGHAI JIAOTONG UNIV

AI summary

Described herein is a method for optimizing a scalable GPU virtualization, comprising: providing each vGPU with a private shadow graphics translation table (GTT); copying vGPU's private shadow GTT to physical GTT along with the context switch, wherein the private shadow GTT allows vGPUs to share an overlapped range of a global graphics memory space.