Dynamic Resource Scheduling for Large-Model Resource Contention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large model applications in a system consume excessive resources, leading to abnormal or inefficient operation of other applications due to unreasonable resource allocation.
Innovation Solution
A resource scheduling method that determines a scheduling mode based on dynamic running information of target applications, adjusting resource utilization to ensure adequate resources for the target application, thereby preventing abnormal or inefficient operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large model application utilizes computing resources in the system, then the large model application can run, but other applications may run abnormally or inefficiently due to excessive resource consumption
Solution Approach 1:
The patent implements dynamic resource scheduling by continuously monitoring resource usage and adjusting the scheduling mode based on real-time system state. The scheduler transitions between different scheduling modes (e.g., from large model priority to general application priority) dynamically, ensuring that resource allocation adapts to current needs and prevents any single application from monopolizing resources to the detriment of others.
Solution Approach 2:
The system changes scheduling parameters based on detected resource usage thresholds. When resource consumption exceeds predefined thresholds, the system modifies scheduling parameters to limit the large model application's resource usage and prioritize other applications, thereby maintaining system-wide reliability while controlling resource consumption.
2Reliability
If resource allocation is optimized for the target application, then the target application operates normally, but other applications may suffer from insufficient resources
Solution Approach 1:
The patent applies different scheduling policies to different applications based on their specific needs and system state. Instead of a uniform scheduling approach, the system implements local optimization by adjusting resource allocation for the target application versus other applications according to their respective requirements, ensuring each receives appropriate resources for its operational context.
Solution Approach 2:
The resource allocation is dynamically adjusted based on system conditions. When the target application requires more resources, the system temporarily prioritizes it; when other applications need resources, the system shifts allocation accordingly. This dynamic balancing act ensures operational stability for the target application while preventing severe resource starvation of other applications.
3Device complexity
If the system uses a fixed resource allocation strategy, then resource management is simple, but it cannot adapt to changing runtime conditions of applications
Solution Approach 1:
The scheduling mechanism transitions from static to dynamic, automatically adjusting resource allocation based on runtime conditions such as application priority, resource usage patterns, and system load. This dynamic approach increases adaptability while managing complexity through automated decision-making algorithms that respond to changing conditions without requiring manual reconfiguration.
Solution Approach 2:
The system implements feedback loops that continuously monitor application performance and resource usage, then use this information to adjust scheduling decisions. This feedback mechanism enables the system to adapt to runtime conditions automatically, improving versatility while keeping the scheduling logic manageable through rule-based or algorithmic responses to detected conditions.
Data Source
AI summary
A resource scheduling method is provided in the present disclosure. The resource scheduling method includes determining a scheduling mode corresponding to a system at least according to dynamic running information of a target application in the system, where the dynamic running information of the target application at least includes one of running information and resource utilization information of the target application; and the scheduling mode characterizes a utilization strategy of a target model for multiple types of resources in the system; and further include based on the scheduling mode, adjusting utilization of at least one type of resource in the system by at least one target model running in the system, such that a corresponding type of resource needed to run the target application is adjusted.