VM Workload Prediction for Server Load Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data center management methods are inefficient due to unpredictable results when applying management programs and potential server overload from time-based VM migrations, leading to high energy consumption and operational costs.
Innovation Solution
A virtual machine (VM) management method that predicts workload using a prediction device, classifying storage loads into groups and forming prediction models to calculate predicted loads, and a VM deployment method that calculates deployment schedules to minimize load variations and server differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If VMs are migrated based on a specific time point, then migration scheduling is simplified, but server overload occurs and data center efficiency deteriorates
Solution Approach 1:
The patent changes the migration trigger from a fixed time-point parameter to a dynamic workload-threshold parameter. The management server continuously monitors VM workload metrics (CPU usage, memory consumption, storage I/O) and initiates migration when workload exceeds predefined thresholds, rather than migrating at predetermined time intervals. This parameter change resolves the contradiction by maintaining simple automated scheduling while preventing server overload through adaptive, condition-based migration decisions.
2Ease of operation
If management programs are applied to data centers, then operational control is improved, but unpredictable results occur due to inability to predict workload outcomes
Solution Approach 1:
The patent implements preliminary action by performing workload prediction and analysis before executing migration operations. The management server collects historical workload data, analyzes patterns using prediction algorithms, and simulates migration outcomes before actually migrating VMs. This preliminary assessment ensures that migration decisions are based on predicted future workload states rather than reactive responses, improving both operational control and result predictability.
Solution Approach 2:
The patent establishes a feedback loop where the management server continuously monitors actual workload metrics, compares them with predicted values, and uses this feedback to refine migration thresholds and prediction models. The system adjusts migration decisions based on feedback from previous migrations and actual workload behavior, thereby improving prediction accuracy and operational reliability over time while maintaining ease of automated operation.
3Reliability
If servers are operated conservatively with stability as top priority, then system stability is maintained, but energy consumption and operational costs increase
Solution Approach 1:
The patent applies dynamics by transitioning from static, conservative server operation to dynamic, adaptive workload management. The system continuously adjusts VM migration decisions based on real-time workload conditions, predicting future workload states and proactively migrating VMs before overload occurs. This dynamic approach maintains system stability through automated load balancing while enabling servers to operate at optimal utilization levels, thereby reducing energy consumption compared to conservative operation where servers run at lower, less efficient capacity.
Data Source
AI summary
A virtual machine (VM) management method may involve simulating a change in deployment of VMs deployed on physical servers including a first physical server and a second physical server physically separated from the first physical server. Server workload prediction based on possible VM deployment may be used for scheduling deployment of VMs to obtain a satisfactory deployment of the VMs on the servers. This may involve classification of storage loads of the VMs, forming a prediction model for predicting prediction load based on storage loads in a classification, and selecting a prediction model for use in determining a target prediction load on a physical server as a result of target VM being analyzed.