Microvisor Emulated Physical Function for Multi-User Resource Sharing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional information handling systems face challenges in sharing resources like GPUs and other accelerator devices between Logically Composed Systems (LCSs) due to their design for single-user interfaces with well-defined APIs that do not support multi-user requirements, leading to the need for custom APIs/OS libraries that are difficult to maintain and update.
Innovation Solution
An Information Handling System (IHS) that includes a microvisor processing system and memory system, which provides an emulated Physical Function (ePF) to enable resource sharing by configuring a microvisor engine to receive requests from an LCS API subsystem, identify resource devices, establish communication channels, and provide resource functionality through these channels, allowing conventional resource device APIs to be used by multiple LCSs without modification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If custom APIs/OS libraries are developed to support multi-user requirements for GPU/accelerator devices, then resource sharing between LCSs is enabled, but system complexity and maintenance burden increase
Solution Approach 1:
The patent introduces a virtualization layer that acts as an intermediary between the GPU/accelerator device and multiple LCSs. This layer presents a standardized interface to LCSs while managing the actual resource allocation and communication with the physical device, thereby enabling multi-user support without requiring each LCS to handle complex custom APIs directly.
Solution Approach 2:
The patent creates virtual copies or representations of the GPU/accelerator device that can be allocated to different LCSs. These virtual instances provide the appearance of direct access while actually routing through the virtualization layer, allowing multiple LCSs to use conventional APIs as if they had direct access to dedicated hardware.
2Adaptability or versatility
If custom APIs/OS libraries are developed for GPU/accelerator devices, then resource sharing is enabled, but ease of maintenance and updates deteriorates
Solution Approach 1:
The virtualization layer serves as a maintainable intermediary that centralizes the multi-user management logic. Updates and maintenance can be performed on this layer without affecting the LCSs or the underlying GPU device, as LCSs continue to use stable conventional APIs while the virtualization layer handles all multi-user coordination.
Solution Approach 2:
The patent segments the system into distinct layers: the LCS layer using conventional APIs, the virtualization layer handling multi-user management, and the physical GPU layer. This segmentation allows independent maintenance of each layer, with the virtualization layer being updated without requiring changes to LCS applications or custom API libraries.
3Productivity
If GPUs/accelerator devices are shared between LCSs, then resource utilization efficiency improves, but compatibility with conventional APIs deteriorates
Solution Approach 1:
The patent creates virtual instances of the GPU device that can be allocated to different LCSs. Each virtual instance appears as a dedicated device to the LCS, allowing it to use conventional APIs without modification. The virtualization layer manages the actual sharing of the physical GPU resource across multiple virtual instances, maintaining API compatibility while enabling efficient resource utilization.
4Ease of operation
If direct access to GPU/accelerator devices is provided to LCSs, then ease of operation improves, but resource sharing capability deteriorates
Solution Approach 1:
The virtualization layer acts as an transparent intermediary that allows LCSs to operate as if they have direct access to GPU devices. It manages resource allocation, isolation, and sharing in the background, while LCSs continue to use familiar direct access patterns with conventional APIs, thus maintaining ease of operation while enabling multi-user sharing.
Data Source
AI summary
An LCS ePF-enabled resource system includes a resource system coupled to a resource management system and resource device(s) configured to provide resource functionality. The resource system includes a microvisor subsystem that the resource management system configures to provide an LCS and an ePF that is presented to the LCS as providing the resource functionality. When the microvisor subsystem receives an LCS request for the resource functionality from an LCS API subsystem in the LCS via the ePF, it transmits a microvisor request to the resource management system for the resource functionality that causes the resource management system to identify the resource device(s) for providing the resource functionality. Based on the identification of the resource device(s), the microvisor subsystem establishes a communication channel with each resource device, and provides the resource functionality to the LCS using the resource device(s) and via each communication channel, the ePF, and the LCS API subsystem.