PCI Resource Allocation for GPU Servers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing resource allocation methods for PCI devices during the BIOS startup process are inadequate, particularly due to limited IO resources, leading to operational issues when the number of PCI devices exceeds 16, and unnecessary allocation of IO resources when connected to a GPU server.
Innovation Solution
A method and device that acquire information from a Switch chip during PCI enumeration, determining if a PCI device is connected to a GPU server to either cancel or allocate IO resources based on a preset rule, ensuring resource availability and normal operation of all PCI devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If each PCI device is allocated with 4k IO resource during BIOS startup, then each device can operate independently, but the total IO resource limit of 64k is exceeded when 16 or more devices are present
Solution Approach 1:
The patent applies local quality by differentiating resource allocation based on device type and connection context. Devices connected to GPU servers receive different allocation (cancelled or reduced IO resources) compared to standalone devices. This selective approach ensures that critical devices maintain full functionality while reducing overall resource consumption to fit within the 64k limit.
Solution Approach 2:
The patent changes the allocation parameters dynamically based on detected device characteristics. By identifying whether a device is connected to a GPU server through enumeration operations, the system adjusts the IO resource allocation parameter from the standard 4k to either 0k or a reduced amount, thereby resolving the contradiction between maintaining device operability and staying within resource limits.
2Reliability
If IO resources are allocated to all PCI devices, then device availability is ensured, but resources are wasted when devices are connected to GPU servers where they are unnecessary
Solution Approach 1:
The patent implements self-service by enabling the system to automatically detect device contexts and make intelligent allocation decisions without external intervention. Through PCI enumeration operations, the system identifies devices connected to GPU servers and autonomously adjusts their resource allocation, eliminating waste while maintaining availability for devices that require it.
Solution Approach 2:
The patent uses feedback mechanisms by performing PCI enumeration operations to gather information about device connections and characteristics. This feedback loop allows the system to continuously monitor device states and adjust resource allocation accordingly, ensuring that IO resources are allocated only when necessary and not wasted on devices connected to GPU servers where they are unnecessary.
3Adaptability or versatility
If the number of PCI devices increases to 16 or more, then system expandability is improved, but device operation fails due to insufficient IO resources
Solution Approach 1:
The patent applies dynamics by making the resource allocation system adaptive and flexible rather than static. The allocation scheme dynamically adjusts based on the number of devices and their connection characteristics, allowing the system to expand to 16 or more devices while maintaining operational stability through intelligent, context-aware resource distribution.
Data Source
AI summary
A method, a device and a medium for allocating a resource based on a type of a PCI device are provided. In a case of running a BIOS program during a start-up process, information of a Switch chip captured by a PCI enumeration operation is acquired. It is determined whether the PCI device is connected to a GPU server based on the information of the Switch chip. An operation of allocating the PCI device with an IO resource is cancelled in a case that the PCI device is connected to the GPU server, and the PCI device is allocated with an IO resource and a memory resource based on a preset allocation rule in a case that the PCI device is not connected to the GPU server.