Equivalence Class Prediction for Server Workload Assignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current task assignment solutions for computational tasks are inefficient and inaccurate, leading to underutilization of servers and increased expenses due to wasted resources, resulting in user dissatisfaction.
Innovation Solution
An equivalence-class-based resource usage prediction system that groups workloads with similar characteristics into clusters, using machine learning to predict resource usage and assign tasks to appropriate servers, reducing the need for individual evaluation of each workload.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If current per-task assignment solutions are used to analyze and predict resource usage for each computational task individually, then task assignment can be performed, but the system becomes inefficient and inaccurate due to the large number of tasks and wide variety of computational tasks
Solution Approach 1:
The patent groups similar computational tasks into equivalence classes based on their resource usage characteristics. Instead of evaluating each task individually, tasks with similar resource patterns are merged into the same equivalence class, allowing the system to predict resource usage for entire classes rather than individual tasks. This merging approach maintains prediction accuracy while dramatically improving assignment efficiency.
Solution Approach 2:
The patent transforms the prediction approach by changing from task-specific parameters to equivalence class-level parameters. By defining tasks through their resource usage patterns (CPU, memory, I/O characteristics) and grouping them by similarity, the system changes the granularity of prediction from individual task details to aggregated class characteristics, improving both efficiency and accuracy.
2Loss of energy
If multiple applications and services are put together in each server to improve resource utilization, then resource efficiency improves, but careful scheduling and predictions are required to ensure each server has enough resources to sustain the applications
Solution Approach 1:
The patent merges the scheduling complexity by grouping multiple applications and services into equivalence classes based on their resource usage patterns. Instead of scheduling each application individually, the system schedules equivalence classes as units, reducing the computational complexity of scheduling while maintaining effective resource allocation across multiple applications on each server.
Solution Approach 2:
The equivalence class prediction module serves multiple functions: it predicts resource usage for scheduling decisions, identifies suitable servers for task assignment, and provides insights for capacity planning. This universal approach handles diverse scheduling requirements through a unified mechanism, reducing overall system complexity.
3Measurement precision
If per-task analysis is performed for each computational task, then individual task requirements can be identified, but the large number of tasks makes the solution inefficient and leads to server under-utilization
Solution Approach 1:
The patent combines individual task analysis into equivalence class-level analysis. By merging tasks with similar resource usage patterns into the same class, the system maintains precise identification of task requirements at the class level while processing assignments much faster than individual task analysis would allow.
Solution Approach 2:
Instead of performing complete individual analysis for every task, the system performs partial analysis by categorizing tasks into equivalence classes based on key resource usage characteristics. This partial action approach captures the essential requirements needed for effective scheduling without the excessive computational cost of full individual task analysis.
Data Source
AI summary
A digital environment includes multiple computing nodes and a scheduling system that assigns workloads to computing nodes. The scheduling system includes an equivalence-class-based resource usage prediction system that receives a workload request and predicts an equivalence class for that workload request based on resource usage over time by the workload request or metadata associated with the workload request. The scheduling system also includes a workload assignment system that assigns the workload request to one or more of the computing nodes based on the predicted equivalence class. The number of equivalence classes is small relative to the total number of workloads that are scheduled (as an example, 10 to 15 equivalence classes for a total number of workloads in the tens or hundreds of thousands).