Distributed ML RAN Execution with Edge Datacenters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radio access networks (RANs) face challenges in efficiently managing resources and deploying applications due to limitations in scalability, fault tolerance, and real-time processing capabilities, particularly in virtualized RANs with distributed edge datacenters.
Innovation Solution
A system comprising far-edge and near-edge datacenters with dedicated computing resources for RAN functions and core network functions, respectively, along with a central controller that optimizes resource allocation and application deployment using a utility function and Bayesian optimizer to maximize efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If dedicated processing hardware is deployed for each base station in a virtualized RAN, then real-time processing capability and reliability are improved, but device complexity and deployment cost increase
Solution Approach 1:
The patent segments the RAN processing functions by deploying separate inference engines at edge data centers for different network functions (RAN functions, core network functions, RIC applications). This segmentation allows each inference engine to be independently managed and deployed on appropriate hardware platforms, improving reliability through distribution while avoiding the complexity of a monolithic system.
Solution Approach 2:
The patent introduces edge data centers as intermediary components between radio units and the core network. These edge data centers host inference engines that perform AI/ML processing, acting as mediators that enable real-time processing without requiring dedicated hardware at every base station, thus improving reliability while managing system complexity.
2Adaptability or versatility
If generic computing resources are used in edge data centers for RAN processing, then scalability and cost-efficiency are improved, but real-time processing capability deteriorates
Solution Approach 1:
The patent applies local quality by deploying inference engines with specialized hardware accelerators (GPUs, FPGAs, ASICs) at edge data centers closer to the radio access network, while using generic computing resources at centralized data centers. This allows real-time processing where needed while maintaining scalability through generic resources elsewhere, resolving the contradiction between processing speed and scalability.
Solution Approach 2:
The patent introduces a spatial dimension to resource allocation by distributing inference engines across multiple edge data centers geographically closer to radio units. This dimensional change enables parallel processing across multiple locations, achieving both real-time performance and scalability simultaneously by processing data locally rather than centralizing all computations.
3Productivity
If multiple AI/ML applications are deployed across distributed edge data centers, then processing capability and accuracy are improved, but resource allocation complexity and coordination overhead increase
Solution Approach 1:
The patent implements feedback mechanisms where the orchestrator receives performance metrics and resource utilization data from distributed inference engines, then dynamically adjusts resource allocation and application deployment decisions. This feedback loop enables the system to coordinate multiple AI/ML applications across edge data centers efficiently, improving processing capability while managing allocation complexity through adaptive control.
Data Source
AI summary
A system, method, and computer-readable media for executing applications for radio interface controller (RIC) management are disclosed. The system includes far-edge datacenters configured to execute a radio access network (RAN) function and a real-time RIC; near-edge datacenters configured to execute a core network function and a near-real-time RIC or a non-real-time RIC; and a central controller. The central controller is configured to: receive inputs of application requirements, hardware constraints, and a capacity of first and second computing resources at the far-edge datacenters and near-edge datacenters; enumerate a plurality of feasible combinations of application locations and configurations that satisfy the application requirements and hardware constraints; incrementally allocate a quant of the first or second computing resources to a feasible combination that would produce a greatest utility from the quant based on a utility function; and deploy each of the plurality of applications.