VM Migration Host Selection via Predictive CPU Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current virtual computing systems face challenges such as CPU contention, which leads to increased VM workload completion times and reduced network bandwidth due to inefficient resource allocation and excessive computations for target host selection during VM migration.
Innovation Solution
The system employs machine learning to predict future CPU usage based on historical data, selecting candidate hosts with lower CPU usage and aggregating their usage with the VM to be migrated, thereby reducing CPU contention and computations required for target host selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional VM migration target host selection is used, then VM migration can be performed, but excessive computations are required leading to reduced network bandwidth and increased completion times
Solution Approach 1:
The system performs preliminary actions by pre-calculating and storing CPU usage metrics for all host devices before migration decisions are needed. The cluster controller continuously monitors and maintains up-to-date CPU usage data for each host, so when a VM migration is triggered by contention, the selection can be made immediately using pre-computed projected CPU usages without performing excessive computations at migration time.
2Reliability
If comprehensive host evaluation is performed for VM migration, then optimal target host selection is achieved, but the number of computations increases leading to slower processing
Solution Approach 1:
The host selection process is segmented into distinct phases: (1) filtering host devices based on current CPU usage thresholds to identify candidate hosts, and (2) computing projected CPU usages only for these filtered candidates. This segmentation reduces the number of hosts requiring comprehensive evaluation while maintaining selection accuracy by ensuring only suitable candidates are considered.
Solution Approach 2:
The system performs partial evaluation by computing projected CPU usages only for candidate hosts that pass the initial filtering criterion (current CPU usage below threshold), rather than evaluating all possible hosts. This partial action approach maintains reliability for suitable hosts while avoiding unnecessary computations for obviously unsuitable candidates.
3Productivity
If CPU contention is not addressed proactively, then system operations continue normally, but VM workload completion times increase and resource allocation efficiency decreases
Solution Approach 1:
The system implements feedback mechanisms by continuously monitoring CPU usage on host devices and detecting contention conditions. When contention is detected on a source host, the cluster controller receives this indication and triggers migration decisions based on real-time CPU usage data, creating a closed-loop control system that proactively manages resource allocation efficiency.
Data Source
AI summary
An illustrative embodiment disclosed herein is a cluster including a plurality of host devices. The plurality of host devices include a first host device. The cluster further includes a cluster controller that receives indication of a contention on the first host device. The cluster controller selects a plurality of candidate host devices among the plurality of host devices based on a plurality of host CPU usages of the plurality of candidate host devices. The cluster controller computes a plurality of projected CPU usages of the plurality of candidate host devices. The cluster controller selects a target host device among the plurality of candidate host devices based on the plurality of projected CPU usages of the plurality of candidate host devices.