Cluster CPU Pressure Scheduling for Fewer VM Live Migrations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cluster distributed resource scheduling methods face issues with frequent live migration of virtual machines due to fluctuating CPU utilization, leading to data security risks and high network resource pressure.
Innovation Solution
A method involving quantization operations on host and vCPU utilization rates, integration of historical pressure values, and live migration based on calculated target CPU pressure values to stabilize CPU utilization and reduce the frequency of live migrations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If real-time CPU utilization rates are used for scheduling, then virtual machines with high CPU loads can be migrated timely, but frequent live migration occurs due to periodic fluctuations caused by timed tasks
Solution Approach 1:
The patent introduces a smoothing window mechanism that periodically aggregates CPU utilization data over time intervals. Instead of reacting to instantaneous CPU spikes caused by timed tasks, the system calculates average CPU utilization across multiple sampling points within a smoothing window, thereby filtering out periodic fluctuations and triggering migration only when sustained high utilization is detected.
2Productivity
If real-time CPU utilization rates are used for scheduling, then resource allocation is responsive, but network resource pressure increases due to frequent live migration
Solution Approach 1:
The system employs periodic sampling of CPU utilization rates combined with a smoothing window that aggregates data over time intervals. This periodic approach with temporal aggregation reduces the frequency of migration decisions by filtering out short-term fluctuations, thereby decreasing unnecessary live migration operations and reducing network resource consumption while maintaining adequate resource allocation responsiveness.
3Stability of the object's composition
If live migration is performed frequently to balance CPU utilization, then resource balance is improved, but memory data loss risk increases
Solution Approach 1:
The patent implements periodic sampling of CPU utilization rates over time intervals and uses a smoothing window to aggregate these samples. This temporal aggregation filters out periodic fluctuations caused by timed tasks, allowing the system to distinguish between transient CPU spikes and sustained high utilization. Migration is triggered only when smoothed CPU utilization exceeds thresholds, reducing unnecessary migrations and associated memory data loss risks while maintaining CPU utilization balance.
4Measurement precision
If instantaneous CPU utilization rates are used, then scheduling is responsive to current load, but periodic fluctuations cause unnecessary migration
Solution Approach 1:
The system uses periodic sampling of CPU utilization rates combined with a smoothing window that aggregates multiple samples over time intervals. This approach maintains measurement precision by continuously sampling CPU load while improving migration operation efficiency by filtering out periodic fluctuations through temporal aggregation. Migration decisions are based on smoothed CPU utilization trends rather than instantaneous values, reducing unnecessary migrations caused by timed task fluctuations.
Data Source
AI summary
A cluster distributed resource scheduling method, apparatus and device, and a storage medium. The method includes: respectively acquiring host CPU utilization rates and vCPU utilization rates; respectively performing a quantization operation on the host CPU utilization rates and the vCPU utilization rates, to obtain host quantized pressure values and virtual machine quantized pressure values; acquiring historical host pressure values and historical virtual machine pressure values; according to the host quantified pressure values and historical host pressure values respectively corresponding to hosts, and virtual machine quantified pressure values and historical virtual machine pressure values respectively corresponding to virtual machines in the hosts, calculating target CPU pressure values respectively corresponding to the hosts; and performing a live migration operation on the virtual machines in the hosts according to the target CPU pressure values.