VM Migration Controller Using ML Outlier Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current virtual computing systems face CPU contention issues, leading to increased VM workload completion times and network bandwidth reduction due to inefficient VM migration processes, which result in excessive resource utilization and computational overhead.
Innovation Solution
Implementing a controller virtual machine that uses machine learning to predict future CPU usage, identifies outlier VMs, assigns weights based on CPU usage thresholds, and iteratively selects VMs for migration to optimize resource allocation and reduce CPU contention, thereby improving network bandwidth and resource efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional VM migration processes are used to balance CPU workload, then CPU contention is reduced, but network bandwidth consumption increases and resource efficiency decreases
Solution Approach 1:
The system performs preliminary identification of outlier VMs using machine learning predictions before migration decisions are made. By pre-computing CPU usage predictions and identifying candidates that need migration, the system avoids reactive migration triggers that consume excessive network bandwidth, thus resolving the contradiction between reducing CPU contention and minimizing network bandwidth consumption.
Solution Approach 2:
The system changes the parameter of VM selection from traditional threshold-based methods to machine learning-based outlier detection. This parameter change enables more precise identification of VMs that actually need migration, reducing unnecessary migrations and thereby decreasing network bandwidth consumption while maintaining effective CPU contention reduction.
2Productivity
If multiple VMs are migrated frequently to balance workload, then resource allocation improves, but computational overhead increases
Solution Approach 1:
The system extracts and focuses only on outlier VMs that deviate significantly from normal CPU usage patterns, rather than evaluating all VMs for migration. This extraction approach reduces the computational overhead by limiting the scope of analysis to only those VMs that truly need migration, while still improving resource allocation efficiency through targeted migrations.
Solution Approach 2:
The system applies partial action by selecting only a subset of VMs (outliers) for migration based on machine learning predictions, rather than migrating multiple VMs frequently. This partial approach reduces computational overhead by avoiding unnecessary evaluations and migrations, while maintaining productive resource allocation through precise targeting of migration candidates.
3Ease of manufacture
If threshold-based VM selection is used for migration, then implementation is simple, but CPU contention reduction effectiveness decreases
Solution Approach 1:
The system introduces machine learning models as an intermediary between the simple threshold-based selection and the actual migration decision. This intermediary layer maintains implementation simplicity by building upon existing threshold concepts while significantly improving CPU contention reduction effectiveness through accurate prediction of future CPU usage patterns and identification of true outlier VMs.
Data Source
AI summary
An illustrative embodiment disclosed herein is a host device including a plurality of virtual machines and a controller virtual machine configured to compute a plurality of central processing unit (CPU) usages corresponding to the plurality of virtual machines. The controller virtual machine is further configured to compute a total usage as a sum of the plurality of CPU usages and to flag one or more outlier virtual machines of the plurality of virtual machines responsive to one or more exceeding CPU usages of the one or more outlier virtual machines being greater than a threshold usage. The controller virtual machine is further configured to assign weights to the one or more outlier virtual machines and to select, for virtual machine migration, a first outlier virtual machine of the one or more outlier virtual machines responsive to the total usage being greater than a target usage.