Cloud Workload Resource Allocation via RNN Probability Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cloud computing systems fail to accurately predict future workload data due to their inability to account for randomness and dynamic fluctuations, leading to resource allocation mismatches that result in either under-allocation causing low service performance or over-allocation leading to high costs, especially for applications with little historical data.
Innovation Solution
The use of recurrent neural networks (RNN) to predict workload probability parameters, which learn nonlinear time-space relationships in future workload data, enabling dynamic resource allocation that satisfies quality of service while minimizing waste and operational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If fixed threshold values are used for resource allocation predictions, then the system is simple to implement, but it cannot respond effectively to dynamic fluctuations in workload
Solution Approach 1:
The patent applies dynamics by transitioning from static fixed threshold values to dynamic probability parameter predictions that adapt to changing workload conditions. The recurrent neural network continuously learns from historical data and updates predictions in real-time, allowing the system to respond to dynamic fluctuations while maintaining manageable complexity through automated adaptation.
Solution Approach 2:
The system implements feedback by using historical workload data to train the recurrent neural network, which then generates probability parameter predictions. These predictions are continuously refined based on actual workload outcomes, creating a closed-loop system that improves accuracy over time while adapting to new patterns in workload behavior.
2Reliability
If resource allocation is based on inaccurate predictions, then operational costs increase due to over-allocation, but service performance decreases due to under-allocation
Solution Approach 1:
The patent changes the prediction parameter from simple threshold values to probability parameters (mean and standard deviation) that capture the statistical characteristics of workload. This parameter transformation enables more nuanced resource allocation decisions, allowing the system to optimize the balance between service performance and operational costs by understanding the distribution and variability of future workload demands.
Solution Approach 2:
The system replaces traditional mechanical rule-based threshold comparisons with an intelligent recurrent neural network that performs probabilistic predictions. This substitution enables the system to handle complexity and uncertainty in workload patterns more effectively, achieving better resource allocation that simultaneously improves service performance and reduces operational costs.
3Measurement precision
If long period historical data is required for predictions, then prediction accuracy improves, but the system cannot serve applications with little or no historical data
Solution Approach 1:
The patent applies preliminary action by pre-training the recurrent neural network on historical workload data from similar applications or general workload patterns. This pre-training establishes baseline prediction capabilities that can be quickly adapted to new applications with little or no historical data, allowing the system to provide accurate predictions from the outset while maintaining the ability to improve accuracy as more application-specific data becomes available.
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for allocating resources based on predictions of workload probability parameters. The method can include collecting a first set of historical workload data generated by operating a first set of one or more applications at a first number of past time instances; predicting probability parameters of a second set of future workload data for operating a second set of one or more applications at a second number of future time instances; and determining future resources allocated to operating the second set of one or more applications for the second number of future time instances, based on allocated current resources, a lower bound of resources to satisfy a quality of service (QoS) for operating the second set of one or more applications, an upper bound of resources to satisfy the QoS, and the predicted probability parameters.