Heterogeneous Task Scheduling Using Global and Submission Time
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing task scheduling algorithms in heterogeneous computing platforms, such as WF2Q, face inefficiencies due to unknown execution time consumption of tasks and high time complexity (O(2*log(N)+N) as they traverse all task channels, leading to unacceptable delays with increasing streams.
Innovation Solution
A method that compares global time and submission time of initial tasks in each stream to determine scheduling, avoiding traversal and search of all task channels, simplifying scheduling and reducing performance overheads by estimating execution times based on historical data and priority considerations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If WF2Q algorithm is used to allocate bandwidth to tasks in different task channels, then fairness and priority policy are achieved, but time complexity increases to O(2*log(N)+N) and scheduling delay becomes unacceptable as number of streams increases
Solution Approach 1:
The patent extracts only the necessary scheduling information (submission time and global time) from each task channel, eliminating the need to traverse and search all task channels. This selective extraction reduces time complexity from O(2*log(N)+N) to O(1) while maintaining scheduling fairness through the time comparison mechanism.
Solution Approach 2:
The patent performs preliminary recording of submission times for all tasks before scheduling decisions are made. By pre-storing this critical information, the scheduling process only needs to compare against the current global time, avoiding the need for real-time traversal and search of task channels during scheduling operations.
2Measurement precision
If traversal and search of all task channels is performed to select task with shortest time consumption, then optimal task selection is achieved, but scheduling time complexity increases to O(2*log(N)+N)
Solution Approach 1:
The patent extracts only the submission time parameter from each task channel, ignoring other parameters like execution time consumption that require complex search and comparison. This extraction simplifies the scheduling algorithm to a direct time comparison operation while maintaining accurate task selection based on submission order.
Solution Approach 2:
Instead of searching for the task with the shortest execution time (forward approach), the patent inverts the approach by selecting tasks based on the earliest submission time. This inversion eliminates the need for complex search algorithms while achieving fair and efficient task selection through simple time comparison.
3Productivity
If execution time consumption of tasks is determined before scheduling, then optimal scheduling decisions can be made, but programming model constraints and performance requirements cannot be met
Solution Approach 1:
The patent performs preliminary recording of submission times without requiring determination of execution time consumption. This preliminary action captures the essential scheduling information needed for fair task selection while avoiding the need to predict or measure execution time, thus maintaining compatibility with programming model constraints.
Solution Approach 2:
The patent introduces global time as an intermediary parameter that mediates between multiple task channels without requiring knowledge of individual task execution times. This intermediary mechanism enables scheduling decisions to be made based on submission order alone, maintaining versatility across different programming models and performance requirements.
Data Source
AI summary
The present disclosure relates to a method for task scheduling and related products, where the related products include a device and a computer-readable storage medium. The device may be included in a computing processing apparatus of a combined processing apparatus, which may include one or more data processing apparatuses. The aforementioned combined processing apparatus may also include an interface apparatus and other processing apparatus. The computing processing apparatus interacts with other processing apparatus to jointly complete user specified computing operations. The combined processing apparatus may also include a storage apparatus, which is connected to the device and other processing apparatuses, respectively, to store data from the device and other processing apparatuses. The solution of the present disclosure may optimize scheduling operations and effectively overcome the shortcomings of existing scheduling strategies.