Dynamic vGPU Migration for Host Capacity and Contention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for assigning virtual graphics processing units (vGPUs) to physical GPUs in computing systems lead to resource fragmentation and hot spots, resulting in underutilization of resources and performance contention.
Innovation Solution
Implement dynamic scheduling techniques to re-arrange vGPU placements based on profiles, allowing for migration and reconfiguration of physical resources to optimize hosting capacity and mitigate resource contention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If breadth-first search technique is applied to assign vGPUs to pGPUs, then VMs are spread across nodes to maximize performance, but resources become fragmented and pGPUs are underutilized
Solution Approach 1:
The patent implements dynamic vGPU migration capabilities that allow the system to adapt placement decisions based on current resource utilization states. The system continuously monitors node resource usage and can migrate vGPUs between nodes to optimize both performance distribution and resource consolidation, resolving the static trade-off between spreading and consolidating approaches
Solution Approach 2:
The system changes the placement parameters dynamically by adjusting the assignment of vGPUs to pGPUs based on monitored resource utilization metrics. When fragmentation is detected, the system modifies placement parameters to consolidate vGPUs onto fewer pGPUs, thereby improving overall resource utilization while maintaining performance goals
2Quantity of substance
If depth-first search technique is applied to assign vGPUs to pGPUs, then the number of pGPUs used is minimized, but nodes become hot spots with resource contention
Solution Approach 1:
The system implements continuous monitoring of node resource utilization and uses this feedback to detect hot spot formation. When resource contention is detected on a node, the system triggers migration of vGPUs from that node to other nodes with available capacity, thereby eliminating hot spots while maintaining the benefits of pGPU consolidation
Solution Approach 2:
The system proactively prevents hot spot formation by monitoring resource utilization trends and initiating vGPU migrations before resource contention becomes severe. By detecting early signs of hot spot development and taking preventive action, the system avoids performance degradation while maintaining consolidated pGPU assignments
3Ease of manufacture
If static placement is applied when vGPU VMs are first powered on, then initial assignment is simple, but the system cannot adapt to changing resource utilization patterns
Solution Approach 1:
The system performs initial static placement of vGPU VMs when they are first powered on, providing a simple baseline assignment. This preliminary action establishes initial resource allocation without complexity, while the system retains the capability to perform subsequent dynamic migrations to adapt to changing conditions
Solution Approach 2:
The system transitions from static initial placement to dynamic ongoing management by implementing continuous monitoring and migration capabilities. The system maintains the simplicity of initial assignment while adding dynamic adaptation through automated vGPU migration based on changing resource utilization patterns
Data Source
AI summary
Various embodiments set forth one or more non-transitory computer-readable media storing program instructions that, when executed by one or more processors, cause the one or more processors to perform steps of determining at least one physical resource of a node in a cluster of nodes is under contention by virtual computing instances or a virtual computing instance cannot be placed on the cluster of nodes, determining a placement for one or more virtual computing instances on the cluster of nodes, each virtual computing instance having a virtual resource associated with a profile that is compatible with a profile associated with a physical resource of a node on which the virtual computing instance is placed, and generating and executing a plan to achieve the placement, the plan including at least one of migrating at least one virtual computing instance or reconfiguring a profile associated with at least one physical resource.