Hierarchical Task Scheduling System for Data Center Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional central task scheduling systems in data centers face bottlenecks due to the high volume of scheduling information that needs to be processed, leading to inefficient task scheduling and failure to meet service level agreements, especially when handling a large number of computing nodes and task requests.
Innovation Solution
A distributed, hierarchical task scheduling system that decentralizes decision-making by introducing cluster and sub-cluster controllers at intermediate levels, allowing only necessary scheduling information to be processed at each level, reducing the load on top-level task schedulers and improving efficiency and scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a central task scheduler processes all scheduling information from all nodes, then comprehensive task scheduling decisions can be made, but the processing load and information overhead increase significantly
Solution Approach 1:
The patent divides the centralized scheduling system into a hierarchical structure with multiple levels: top-level task scheduler, intermediate-level cluster controllers, and bottom-level nodes. Each level processes scheduling information locally, segmenting the overall processing load. Cluster controllers manage specific clusters of nodes, filtering and preprocessing scheduling information before forwarding to the top-level scheduler, thereby reducing the information overhead while maintaining comprehensive scheduling capability.
Solution Approach 2:
The patent introduces an intermediate dimension (cluster level) between the top-level scheduler and individual nodes. This dimensional addition creates a multi-level hierarchy where scheduling decisions are made at appropriate levels, reducing the direct information processing burden on the top-level scheduler while ensuring comprehensive task allocation across the entire system.
2Productivity
If a central task scheduler handles all task requests directly, then centralized control is maintained, but processing speed and scalability decrease
Solution Approach 1:
The patent segments the task scheduling process across multiple hierarchical levels. Cluster controllers at the intermediate level handle task requests for their respective clusters, performing local scheduling decisions and filtering out routine tasks. This segmentation increases processing speed by distributing operations, while the hierarchical structure maintains centralized control for overall system coordination.
Solution Approach 2:
Cluster controllers serve as intermediary components between the top-level task scheduler and individual nodes. These intermediaries preprocess task requests, perform local scheduling optimizations, and filter information before communicating with higher or lower levels. This intermediary layer accelerates processing by handling routine decisions locally while maintaining system-wide coordination.
3Measurement precision
If detailed scheduling information is collected from all nodes, then accurate scheduling decisions can be made, but information overhead and communication load increase
Solution Approach 1:
The patent extracts and processes scheduling information at appropriate hierarchical levels rather than transmitting all detailed information to the top-level scheduler. Cluster controllers collect and process scheduling information from their respective nodes, extracting only the essential and relevant information needed for cluster-level decisions. This extraction maintains scheduling accuracy for local decisions while reducing the volume of information that needs to be transmitted upward in the hierarchy.
Solution Approach 2:
The patent applies local quality by enabling each hierarchical level to process and make scheduling decisions based on locally available information. Cluster controllers maintain and use scheduling information specific to their clusters, making accurate local scheduling decisions without requiring complete system-wide information. This local processing maintains decision accuracy at each level while minimizing information overhead in communications.
Data Source
AI summary
A computing system comprises a task scheduler in a top-level, at least two cluster controllers in an intermediate-level, and a plurality of nodes in a bottom level. A task scheduler is configured to receive a task request, obtain an adapted task request from the received task request, based on scheduling information stored in the task scheduler, and forward the adapted task request to one of the at least two cluster controllers. Each cluster controller is configured to receive the adapted task request, obtain an adapted cluster task request from the received adapted task request based on cluster scheduling information stored in the cluster controller, and directly or indirectly forward the adapted cluster task request to a node in the plurality of nodes.