Workload Scheduler Using Expected Runtime Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In compute grid environments, existing workload scheduling methods are ineffective due to reliance on inaccurate indicators like CPU model and memory, failing to consider additional factors such as database access distance, leading to reduced performance and processing efficiency.
Innovation Solution
A method and device for improved task scheduling that determines task categories and selects CPUs based on stored expected runtimes for standard computation units, using a matrix management module to update and utilize an expected runtime matrix for adaptive scheduling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If CPU model and memory are used as indicators for task dispatching, then the scheduling process is simple, but the accuracy of task execution time prediction is low
Solution Approach 1:
The patent transforms the scheduling approach by changing from using static hardware parameters (CPU model, memory) to using dynamic performance parameters (expected runtime for standard computation units). This allows the system to capture actual execution characteristics while maintaining a manageable scheduling framework.
Solution Approach 2:
The system implements feedback mechanisms by measuring actual task execution times and using these measurements to update expected runtime values for different CPU-task category combinations. This feedback loop continuously improves prediction accuracy without significantly increasing scheduling complexity.
2Productivity
If traditional scheduling methods are used, then the scheduling process is fast, but the processing efficiency and performance are reduced
Solution Approach 1:
The system performs preliminary classification of tasks into categories and pre-calculates expected runtimes for each CPU-task category combination before actual task dispatch. This preliminary preparation enables faster and more accurate scheduling decisions, improving both processing efficiency and reducing overall execution time.
Solution Approach 2:
The patent applies different scheduling strategies and performance metrics tailored to specific task categories and individual CPU characteristics. By optimizing scheduling decisions locally for each task-CPU combination rather than using a universal approach, the system achieves superior overall processing efficiency.
3Adaptability or versatility
If virtual machine hosting is used, then resource consolidation is achieved, but the CPU model no longer accurately represents the actual processing capability
Solution Approach 1:
The system creates a mapping between virtual CPU models and actual physical CPU performance characteristics by measuring and storing expected runtimes. This copying approach allows the scheduler to work with virtual machine abstractions while maintaining accurate knowledge of underlying physical processing capabilities, thus preserving resource consolidation benefits without sacrificing scheduling accuracy.
Data Source
AI summary
Methods, non-transitory computer readable media, and workload scheduler devices that determine one of a plurality of task categories for a received task. A stored expected runtime for each of a plurality of CPUs to execute one standard computation unit (SCU) in the determined one of the plurality of task category is obtained. One of the plurality of CPUs is selected based on the stored expected runtime. The task is dispatched to the selected one of the plurality of CPUs for execution. Accordingly, with this technology, tasks associated with workloads can be more effectively dispatched and more effectively processed by a CPU pool.