Mobile-Cloud Task Scheduling for AI Compute and Power Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The limited computing power on terminal devices, such as mobile phones and tablets, cannot meet the high computing demands of complex deep learning models, leading to poor performance and increased costs when relying solely on cloud computing.
Innovation Solution
A mobile-cloud collaborative computing power dispatching method that dynamically allocates tasks between terminal devices and cloud servers based on factors like task type, power consumption, temperature, network status, and cloud load, using a mobile-cloud collaborative programming framework to balance performance and costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep learning models are made more complex to meet AI computing requirements, then computing performance is improved, but power consumption and heat generation increase beyond terminal device limits
Solution Approach 1:
The patent segments the computing tasks into different parts: complex deep learning models are divided into components that can be executed on cloud servers, while simpler tasks remain on terminal devices. This segmentation allows high-computing-power tasks to be offloaded to the cloud, reducing power consumption and heat generation on terminal devices while maintaining overall computing performance.
Solution Approach 2:
The patent introduces a mobile-cloud collaborative computing framework as an intermediary between terminal devices and cloud servers. This framework enables seamless task distribution and result aggregation, allowing terminal devices to leverage cloud computing power without directly bearing the full computational load, thus resolving the contradiction between performance requirements and power consumption limits.
2Productivity
If all tasks are executed on cloud servers to meet high computing requirements, then computing performance is improved, but system complexity and network dependency increase
Solution Approach 1:
The patent applies local quality by enabling terminal devices to execute simpler tasks locally while offloading complex tasks to the cloud. This creates a heterogeneous computing architecture where different parts of the system have different capabilities, optimizing the balance between local execution efficiency and cloud computing power utilization, thereby reducing unnecessary network dependency and system complexity.
Solution Approach 2:
The patent implements dynamic task allocation where the decision of whether to execute a task locally or on the cloud is made based on real-time conditions such as network status, device load, and task characteristics. This dynamic approach allows the system to adapt to changing conditions, reducing complexity by making context-aware decisions rather than following a fixed execution model.
3Device complexity
If terminal devices use their own computing power resources, then system simplicity is maintained, but computing power is insufficient for complex AI tasks
Solution Approach 1:
The patent creates a universal mobile-cloud collaborative computing framework that can handle both simple local tasks and complex cloud-based tasks through a unified architecture. This multi-functional system allows terminal devices to seamlessly switch between using local computing power and cloud computing power based on task requirements, maintaining system simplicity while providing access to extensive computing resources when needed.
4Productivity
If cloud computing resources are used extensively, then computing performance is improved, but costs increase
Solution Approach 1:
The patent applies partial action by selectively offloading only the necessary portions of tasks to cloud servers rather than executing all tasks on the cloud. This approach uses cloud computing resources partially - only when and where needed - thereby maintaining computing performance for complex tasks while reducing overall cloud resource consumption and associated costs compared to fully cloud-based execution.
Data Source
Figure 1
Figure 2~3
Figure 4A~4C
AI summary
Embodiments of this application provide a mobile-cloud computing power collaborative dispatching method, a related system, and a device. The method may be applied to a mobile-cloud computing power collaborative dispatching system, and the system may include a terminal device and a cloud server. The terminal device may execute a dispatching policy, to determine to dispatch a task in a program to a mobile side for execution or a cloud side for execution. In this way, a cloud computing power becomes a part of a mobile computing power, implementing collaborative dispatching of mobile and cloud tasks. This solves a problem of an insufficient mobile computing power and avoids an idle mobile computing power and high costs on the cloud side that are caused by placing all logic and computing powers on the cloud side.