Virtual Machine Migration Performance Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual machine management techniques struggle to accurately estimate performance when moving virtual machines between servers with different resource characteristics, such as varying CPU clocks and architectures, leading to inefficiencies in resource allocation and potential overloading.
Innovation Solution
A virtual-machine managing device and method that acquire performance models for each server, convert performance information into workload amount and characteristic value combinations, and estimate performance on destination servers using these models, allowing for accurate resource allocation and efficient virtual machine migration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If performance information is measured using virtual servers with the same OS type and performance-measuring program, then measurement standardization is improved, but measurement precision deteriorates because workload characteristics differ from actual virtual servers
Solution Approach 1:
The patent changes the measurement approach from using standardized virtual servers to measuring actual workloads directly on physical servers. It introduces workload characteristic value extraction that captures specific workload parameters (CPU usage patterns, memory access patterns, I/O characteristics) to enable accurate performance estimation without requiring identical virtual server images.
Solution Approach 2:
Instead of copying virtual server images for measurement, the patent creates a virtual copy of workload characteristics through performance models. These models capture the essential behavior patterns of workloads and allow performance estimation across different physical servers without actually migrating or replicating the full virtual servers.
2Adaptability or versatility
If virtual machines are moved between physical machines with different CPU architectures, then resource utilization flexibility is improved, but performance estimation accuracy deteriorates due to architecture-specific performance variations
Solution Approach 1:
The patent introduces architecture-specific performance models that capture CPU architecture characteristics (clock speed, instruction set, cache hierarchy) as parameters. These models translate workload characteristic values into performance estimates specific to each physical machine's architecture, enabling accurate cross-architecture performance prediction.
Solution Approach 2:
The patent applies local quality by creating performance models tailored to each physical server's specific characteristics rather than using a generic model. Each physical machine has its own performance model that incorporates its unique hardware specifications and performance characteristics, allowing for localized accurate estimation.
3Measurement precision
If performance models are created for each physical server, then performance estimation accuracy is improved, but system complexity increases
Solution Approach 1:
The patent creates a universal performance model framework that can be applied across all physical servers. Instead of developing completely separate models for each server, the system uses a standardized model structure that is instantiated with server-specific parameters, reducing overall system complexity while maintaining accuracy.
Solution Approach 2:
The system automatically generates performance models for each physical server by measuring actual workload performance characteristics on that server. This self-service approach eliminates the need for manual model creation and allows the system to adapt models automatically to changing server conditions.
Data Source
AI summary
Provided is a virtual-machine managing device including: a model acquiring unit that acquires, for each server device, a performance model indicative of plural correspondent relationships between a workload amount and performance information on a workload; a performance-information acquiring unit that acquires the performance information on a virtual machine to be moved running on a current server device; a conversion unit that converts the performance information on the virtual machine to be moved into a combination of the workload amount and the workload characteristic value concerning the virtual machine to be moved, by using the performance model of the current server device; and an estimating unit that estimates performance information on the virtual machine to be moved on a destination server device serving as a candidate for a destination of movement of the virtual machine to be moved, by applying the combination converted by the conversion unit to the performance model of the destination server device.