GPU Memory Pool Manager for Shared Virtual GPU Memory Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In virtualized computing environments, GPUs often experience unequal memory resource utilization leading to GPU occupancy issues, where some GPUs are fully allocated while others remain underutilized, resulting in disparities in VM performance.
Innovation Solution
A GPU memory pool manager (GMP manager) abstracts physical GPU memory resources across multiple GPUs, creating a virtual shared pool (VSGMP) accessible to all VMs, managing memory transactions and allocations to optimize resource utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a 1:1 correspondence between VMs and GPUs is maintained, then GPU assignment simplicity is improved, but GPU memory occupancy equality deteriorates
Solution Approach 1:
The patent merges the memory resources of multiple GPUs into a single shared memory pool that is commonly accessed by multiple VMs. This pooling approach allows VMs to access memory from any GPU in the pool, enabling more flexible resource allocation and improving memory occupancy equality across GPUs while maintaining simple GPU assignment through the hypervisor.
2Speed
If GPU memory is entirely allocated to assigned VMs, then VM memory access performance is improved, but overall system resource utilization deteriorates
Solution Approach 1:
The patent implements dynamic memory allocation within the shared GPU memory pool. The hypervisor can dynamically allocate memory from any GPU in the pool to VMs based on real-time demand, rather than being locked into static 1:1 GPU-VM assignments. This dynamic allocation improves overall system resource utilization while maintaining fast memory access through the shared pool architecture.
Solution Approach 2:
The hypervisor acts as an intermediary between VMs and GPU memory resources. It manages the shared memory pool and mediates memory access requests from multiple VMs, enabling flexible resource sharing while maintaining performance through controlled access paths.
3Adaptability or versatility
If multiple GPUs are deployed with 1:1 VM mapping, then system scalability is improved, but memory resource distribution fairness deteriorates
Solution Approach 1:
The patent creates a universal shared memory pool that serves multiple functions: it can be accessed by any VM regardless of which physical GPU the VM is logically assigned to. This universal access model maintains system scalability through flexible VM deployment while ensuring fair memory resource distribution by allowing memory sharing across all GPUs in the pool.
Data Source
AI summary
Disclosed methods provide a virtualized shared graphics processing unit (GPU) memory pool (VSGMP) to virtual machines running on an information handling system. The VSGMP may be implemented with a GPU memory pool (GMP) manager, featuring logic for abstracting the GPU memory pool from the physical memory resources of two or more GPUs. The GMP manager logic may be supported by a lightweight secure operating system (LSOS) capable of enabling functionality for virtualization and other use cases. Disclosed methods manage GPUs in an information handling system featuring two or more GPUs running virtual machines (VMs). When a GPU is assigned to a VM, disclosed methods perform one or more GPU resource allocation operations that support virtual shared pooling of the physical memory resources of two or more GPUs. In at least some embodiments, the GPU allocation operations include allocating at least some non-memory resources of the GPU exclusively to the applicable VM.