Virtual Machine Rebalancing for Datacenter Processor Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Datacenter servers experience uneven processing loads, leading to 'hot spots' where some processors consume more power than others due to uneven virtual machine (VM) allocation, resulting in inefficiencies and increased wait times for virtual machines.
Innovation Solution
A system that monitors and reallocates VMs across hypervisors based on processor usage and steal data, ensuring that no single processor exceeds a threshold consumption rate and that VMs are distributed evenly to maintain balanced processing loads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If VMs are allocated to hypervisors based on initial distribution, then datacenter can operate with simple allocation logic, but processor consumption becomes uneven creating hot spots
Solution Approach 1:
The patent implements a feedback mechanism where the system continuously monitors processor consumption metrics (CPU usage, memory usage, I/O operations) for each hypervisor and automatically triggers VM reallocation when imbalance thresholds are exceeded. This closed-loop control resolves the contradiction by maintaining simple operational logic while achieving balanced processor consumption through automated adjustments based on real-time feedback.
Solution Approach 2:
The system dynamically adjusts VM allocation based on changing workload conditions by continuously evaluating processor consumption metrics and reallocating VMs from overloaded to underloaded hypervisors. This dynamic approach transforms the static allocation into an adaptive system that maintains balance without requiring complex manual intervention, resolving the contradiction between operational simplicity and consumption uniformity.
2Productivity
If more VMs are assigned to a single hypervisor to increase processing capacity utilization, then fewer hypervisors are needed, but that hypervisor runs hotter and becomes a performance bottleneck
Solution Approach 1:
The system uses feedback control to monitor processor temperature and consumption metrics, automatically redistributing VMs from overheated hypervisors to cooler ones. This maintains high overall processing capacity utilization while preventing any single hypervisor from becoming a thermal bottleneck, resolving the contradiction between productivity and temperature control.
Solution Approach 2:
The patent applies local quality by allowing different hypervisors to handle different workload types or intensities based on their current thermal and performance state. Rather than uniform allocation, the system optimizes each hypervisor's workload composition locally, enabling high aggregate utilization while maintaining acceptable temperature levels across the datacenter infrastructure.
3Device complexity
If VMs are concentrated on fewer hypervisors, then resource allocation is simpler, but VM wait time for processor access increases due to steal time
Solution Approach 1:
The system implements feedback monitoring of VM steal time metrics and automatically triggers reallocation when wait times exceed thresholds. This maintains simple resource allocation logic while reducing VM processor wait time by dynamically balancing the load across hypervisors based on real-time performance feedback, resolving the contradiction between allocation simplicity and time loss.
4Use of energy by moving object
If the system continuously monitors and rebalances VMs to eliminate hot spots, then processor consumption is balanced, but system complexity increases
Solution Approach 1:
The patent employs feedback-based automated control that monitors processor consumption metrics and triggers rebalancing only when imbalance thresholds are exceeded. This approach achieves balanced processor consumption while limiting system complexity by using rule-based automated decisions rather than continuous complex optimization algorithms, making the monitoring and rebalancing process manageable and scalable.
Data Source
AI summary
This application relates to apparatus and methods for reallocating tasks among servers, such as cloud datacenter compute servers. A control server may identify hypervisors running virtual machines that are consuming higher processing power, and may reallocate one or more of the virtual machines to one or more hypervisors consuming lower processing power. In some examples, the control server reallocates virtual machines among hypervisors based on how many virtual machines each hypervisor is running. In some examples, the control server identifies virtual machines associated with higher processing steal times, and may reallocate the identified virtual machines to hypervisors that are consuming lower processing power. The virtual machines may be reallocated to a different hypervisor executing on a same server, to a different hypervisor executing on a different server in a same datacenter, or to a different hypervisor executing on a different server in a different datacenter.