Edge Accelerator Sharing for Resource-Limited Edge Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current edge devices have limited computing resources, including small memory size and low processor capability, which fail to meet the requirements of edge computing applications.
Innovation Solution
A control server installs a daemon on edge devices to monitor accelerators, configures an enabled flag for devices with connected accelerators, and selects edge devices to share accelerators for executing target applications, allowing edge devices to utilize additional computing resources through accelerator sharing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If edge devices are used for edge computing applications, then edge computing can be realized near data sources, but computing resources (memory size and processor capability) are insufficient to meet application requirements
Solution Approach 1:
The patent merges computing resources from multiple edge devices by introducing accelerators that can be shared across devices. The control server manages resource allocation and combines available computing power from different edge devices to create a pooled resource pool that meets application requirements.
Solution Approach 2:
The patent makes accelerators universal resources that can serve multiple edge devices simultaneously. A single accelerator can be dynamically allocated to different edge devices based on demand, allowing one physical resource to fulfill multiple computing functions across the network.
2Productivity
If accelerators are connected to specific edge devices, then computing resources increase, but device complexity increases due to accelerator management and configuration
Solution Approach 1:
The control server acts as an intermediary between edge devices and accelerators. It manages accelerator connections, handles configuration, and coordinates resource allocation, thereby isolating the complexity of accelerator management from individual edge devices and their applications.
Solution Approach 2:
The system implements feedback mechanisms where the control server monitors accelerator status and edge device resource needs, then dynamically adjusts resource allocation. This closed-loop control simplifies management by automatically responding to changing conditions without manual intervention.
Data Source
AI summary
Various embodiments of the teachings herein include an edge computing method. The method may include: installing a daemon on a first edge device connected to a control server, wherein the daemon sends information to the server when an accelerator is connected to the edge device indicating the first edge device is connected to an accelerator; receiving the connection information from the daemon and configuring an accelerator enabled flag for the first edge device indicating the first edge device is connected to an accelerator; receiving a sharing request from either edge device, wherein the sharing request requesting an accelerator execute a target application; labeling an edge device with the accelerator enabled flag as a second edge device according to the sharing request; and loading the target application onto the second edge device so the accelerator connected to the second edge device executes the target application.