Hypervisor Shadow Graphics Translation Table Conflict Check
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current GPU virtualization technologies face scalability issues and performance degradation due to static global graphics memory space partitioning, which limits the number of virtual machines that can be supported and results in insufficient memory for resource-intensive workloads.
Innovation Solution
Implementing a hypervisor with a command parser and GPU scheduler that generate and manage on-demand shadow global graphics translation tables (SGGTTs) to dynamically allocate and translate graphics memory addresses, allowing multiple virtual machines to share a physical GPU without conflicts and optimize memory usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If static global graphics memory space partitioning is used, then memory allocation is simple and stable, but the number of supported virtual machines is limited and memory utilization is insufficient
Solution Approach 1:
The patent implements dynamic memory allocation by replacing static partitioning with on-demand memory allocation. The hypervisor dynamically creates and manages shadow global graphics translation tables (SGGTTs) based on actual workload requirements, allowing the system to adapt to varying memory needs of different virtual machines rather than being constrained by fixed partitions.
Solution Approach 2:
The system changes the memory allocation parameter from fixed static values to dynamic values that adjust based on workload. By implementing conflict checking mechanisms and on-demand SGGTT creation, the memory allocation parameters can be modified in real-time to optimize for the number of supported virtual machines without compromising system stability.
2Productivity
If on-demand shadow global graphics translation tables are implemented, then memory utilization is optimized and more VMs are supported, but system complexity and conflict checking overhead increase
Solution Approach 1:
The patent implements preliminary conflict checking before allocating graphics memory addresses to virtual machines. The hypervisor proactively checks for potential conflicts in the SGGTT before committing memory allocations, preventing future conflicts rather than resolving them after they occur. This preliminary action reduces the overall complexity of conflict management.
Solution Approach 2:
The shadow global graphics translation table serves as an intermediary layer between the virtual machines and the physical graphics memory. This intermediary structure allows the system to manage complex translation and conflict checking in a organized manner, improving memory utilization while containing complexity within the translation layer rather than propagating it throughout the entire system.
3Reliability
If graphics memory addresses are dynamically allocated, then memory conflicts between VMs are reduced, but the overhead of address translation and conflict checking increases
Solution Approach 1:
The system performs preliminary conflict checking and validation when creating and updating shadow global graphics translation tables. By checking for address conflicts before they occur and establishing valid translations in advance, the system reduces the need for time-consuming conflict resolution during runtime, thereby improving reliability without excessive time overhead.
Solution Approach 2:
The patent implements shadow translation tables that copy and mirror the structure of guest physical address spaces. This copying mechanism allows the hypervisor to validate and translate addresses efficiently by comparing against the shadow copy, reducing translation time while maintaining reliable conflict avoidance through the validated shadow structure.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An apparatus may include a graphics processing unit (GPU) and a hypervisor. The hypervisor may include a command parser to parse graphics memory addresses associated with a workload of a virtual machine of the apparatus, and generate a first shadow global graphics translation table (SGGTT) for translating the graphics memory addresses. The hypervisor may further include a GPU scheduler to check conflict between the first SGGTT and a second SGGTT containing graphics memory addresses used by working sets being-executed or to-be-executed by the render engines of the GPU, and schedule the workload of the virtual machine to a render engine when there is no conflict between the first SGGTT and the second SGGTT. Other embodiments may be described and/or claimed.