VM Migration Destination Selection via Probability Distribution Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining a migration destination for virtual machines in public clouds fail to effectively address temporary or spike-like high loads, leading to resource competition due to statistical processing that rounds resource utilization rates, thereby neglecting these transient load conditions in evaluating server migration destinations.
Innovation Solution
An operation management apparatus that generates a continuous probability distribution of resource utilization rates for each virtual machine, creates resource utilization rate estimation data by modeling the probability distribution of resource utilization rates at minute intervals, and calculates competition occurrence probabilities to identify a migration destination server that minimizes resource competition risks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If statistical processing (average value) is used to evaluate server resource utilization, then calculation simplicity is improved, but measurement precision deteriorates due to rounding that neglects temporary or spike-like high loads
Solution Approach 1:
The patent changes the parameter representation from discrete average values to continuous probability distribution functions. Instead of using simple statistical averages that round transient loads, the system models resource utilization rates as continuous probability distributions, capturing temporary and spike-like high loads through the distribution's shape and variance characteristics.
2Device complexity
If migration destination is determined based on average resource utilization rates, then device complexity is reduced, but reliability deteriorates due to resource competition from temporary high loads being overlooked
Solution Approach 1:
The patent introduces feedback mechanisms where the probability distribution models are continuously updated based on observed resource utilization patterns. The system calculates competition occurrence probabilities by comparing the target VM's load distribution with the destination server's existing load distribution, using this feedback to make informed migration decisions that account for temporary high loads.
Solution Approach 2:
The patent performs preliminary analysis by generating probability distribution functions and calculating competition occurrence probabilities before making migration decisions. This preliminary action allows the system to evaluate potential resource competition risks in advance, rather than reacting to average utilization rates after the fact.
3Measurement precision
If continuous probability distribution modeling is used for resource utilization rates, then measurement precision is improved, but device complexity increases due to complex calculations required
Solution Approach 1:
The patent creates simplified representations (copies) of the complex resource utilization patterns through probability distribution functions. Instead of processing raw, high-resolution utilization data directly, the system fits these data to standard probability distribution forms, which can then be manipulated using well-established statistical methods to calculate competition risks.
4Productivity
If migration decisions consider temporary or spike-like high loads, then resource allocation efficiency is improved, but loss of time increases due to more complex evaluation processes
Solution Approach 1:
The patent replaces complex mechanical-style calculations (detailed time-series analysis of utilization rates) with statistical field theory approaches. By modeling resource loads as probability distributions and using analytical solutions for convolution and comparison of these distributions, the system achieves accurate evaluation of temporary high loads without requiring exhaustive computational analysis of historical data.
Data Source
AI summary
An operation management apparatus includes a processor. The processor generates a VM load model for each virtual machine running on an information processing system, generates resource utilization rate estimation data based on VM load models of a virtual machine group running on the physical machine and a VM load model of a first virtual machine, for each of physical machines except for a first physical machine on which the first virtual machine is running, generates a resource competition occurrence model based on the resource utilization rate of the physical machine, calculates a statistical value of competition occurrence probabilities of the resource, for each of the physical machines except for the first physical machine, based on the resource utilization rate estimation data and the resource competition occurrence model, specifies the migration destination physical machine based on the statistical value, and outputs information of a specified migration destination physical machine.