GPU Usage Profiles for vRAN Package Allocation Under Latency Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In a multi-GPU system, the allocation of GPUs to software packages can vary system performance, and there is a need for a more efficient method to support higher performance in a virtualized radio access network (vRAN).
Innovation Solution
A method and apparatus for determining GPU usage profiles based on package information and GPU state information, including computation amount, latency requirements, and power consumption, to optimize GPU allocation among software packages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If GPUs are allocated to multiple software packages using a GPU scheduler, then resource utilization is improved, but system performance varies and cannot be optimized for different package requirements
Solution Approach 1:
The patent applies local quality by creating distinct GPU usage profiles for different software packages (e.g., deep learning packages, image processing packages, video processing packages). Each profile contains package-specific parameters such as computation amount, latency requirements, and power consumption characteristics. This allows the GPU scheduler to allocate resources according to the specific needs of each package rather than using a uniform allocation strategy, thereby maintaining both high resource utilization and consistent system performance across different workloads.
2Productivity
If GPU usage profiles are created for different software packages, then system performance is optimized, but device complexity increases
Solution Approach 1:
The patent implements universality by designing a standardized GPU usage profile structure that can serve multiple software packages with different requirements. The profile template includes universal parameters (computation amount, latency requirements, power consumption) that apply to all packages, while allowing package-specific values to be filled in. This multi-functional profile system simplifies management compared to creating entirely separate allocation mechanisms for each package type, reducing the overall complexity while maintaining optimization capabilities.
Solution Approach 2:
The patent applies parameter changes by dynamically adjusting GPU allocation based on the specific parameters defined in each software package's usage profile. Rather than using fixed allocation rules, the system modifies allocation parameters (such as resource quantity, priority levels, and power limits) according to the package's computation requirements, latency constraints, and power consumption characteristics. This parameter-based approach provides a systematic way to manage complexity through configurable variables rather than hard-coded allocation logic.
3Productivity
If GPU allocation is optimized for specific packages, then package performance is improved, but allocation flexibility decreases
Solution Approach 1:
The patent implements dynamics by making GPU allocation flexible and adaptive rather than static. The GPU scheduler continuously monitors software package execution states and dynamically adjusts resource allocation based on current workload conditions and the defined usage profiles. This dynamic approach allows the system to optimize performance for specific packages while maintaining the ability to adapt to changing requirements, as the allocation can be modified in real-time without requiring rigid pre-configuration for all possible scenarios.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method, performed by a network entity, of executing a software package in a wireless communication system, is provided. The method includes identifying package information of each of a plurality of software packages and graphics processing unit (GPU) state information used by the plurality of software packages with respect to a plurality of user equipments (UEs) connected to the network entity, when a workload of at least one of the plurality of packages is changed, determining to change a GPU usage profile, based on the package information and the GPU state information, determining the number of packages to process the workload, identifications (IDs) of GPUs to be allocated to the packages, and usage locations of the GPUs, based on the package information and the GPU state information, updating the GPU usage profile, according to the determined IDs of the GPUs and the determined usage locations of the GPUs, and allocating the GPUs to the plurality of software packages, based on the updated GPU usage profile.