Intrusiveness-Based Task Scheduling for Federated Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing task scheduling methods in federated learning systems fail to adequately consider the intrusiveness of tasks on heterogeneous and dynamically changing computing devices, leading to inefficient resource utilization and potential performance degradation.
Innovation Solution
Implementing a machine learning-based approach to determine an intrusiveness metric on each computing device, which is used by a scheduler device to optimize task distribution and avoid overloading devices, thereby improving resource utilization and maintaining performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing scheduling methods are used in federated learning environments, then the system can operate with simple task distribution, but computing resources are underutilized or overutilized and device performance is degraded
Solution Approach 1:
The patent implements a feedback mechanism where computing devices report intrusiveness metrics to the scheduler device, which uses this information to dynamically adjust task scheduling decisions. This closed-loop feedback enables the system to learn from actual device performance and optimize resource allocation over time, resolving the contradiction between simple scheduling and effective resource management.
Solution Approach 2:
The system enables computing devices to self-assess their own intrusiveness metrics and autonomously report this information to the scheduler. This self-service approach allows devices to participate actively in the scheduling process without requiring complex centralized control, improving resource utilization while maintaining manageable system complexity.
2Reliability
If tasks are scheduled without considering device performance impact, then scheduling is simpler and faster, but device performance degrades due to underutilization or overutilization
Solution Approach 1:
The patent performs preliminary assessment of intrusiveness metrics before task scheduling occurs. By evaluating device capabilities and potential performance impact in advance, the system can make informed scheduling decisions that prevent performance degradation without requiring complex real-time adjustments during task execution.
Solution Approach 2:
The scheduling system dynamically adapts to changing device conditions by continuously monitoring intrusiveness metrics and adjusting task assignments accordingly. This dynamic approach allows the system to maintain optimal performance by responding to real-time device state changes while avoiding rigid, time-consuming scheduling procedures.
3Adaptability or versatility
If heterogeneous devices with varying capabilities are managed without intrusiveness metrics, then device diversity is accommodated, but task distribution is suboptimal and performance degradation occurs
Solution Approach 1:
The patent applies local quality by tailoring scheduling decisions to individual device characteristics. Each computing device's specific intrusiveness metrics and capabilities are considered when assigning tasks, allowing the system to optimize task distribution for each device's unique properties rather than using a one-size-fits-all approach.
Solution Approach 2:
The system changes scheduling parameters based on device intrusiveness metrics and capabilities. By adjusting task assignment decisions according to measured performance impact and device characteristics, the system optimizes task distribution across heterogeneous devices, improving both adaptability and productivity simultaneously.
Data Source
AI summary
Examples of computing task scheduling based on an intrusiveness metric are described. In an example, an intrusiveness metric that indicates an impact of a computing task on performance of a computing device may be determined with an intrusiveness machine learning model. The intrusiveness metric may be sent to a scheduler device to determine distribution of additional computing tasks according to a scheduling machine learning model.