Cloud Oversubscription System Using V-ARMA for Overload Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Traditional methods for predicting and addressing overload conditions in cloud environments are inadequate due to their reliance on aggregate data, which fails to account for the elasticity of cloud computing and the varying resource utilization patterns of virtual machines, leading to poor prediction and delayed resolution of oversubscription issues.

Innovation Solution

A cloud oversubscription system that models current and future probability of overload on a per-host basis using a Vector Auto Regressive Moving Average (V-ARMA) model, analyzing service level agreement availability values to identify probable overload conditions and recommending actions such as VM termination or migration before actual overload occurs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If traditional aggregate-based overload forecasting is used, then the system can maintain simplicity in monitoring, but the prediction accuracy deteriorates due to inability to account for cloud elasticity and individual VM patterns

Engineering Contradiction:

Improvemonitoring complexityVSAvoidoverload prediction accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the monitoring approach by analyzing each virtual machine's resource utilization patterns individually rather than using aggregate host-level data. This per-VM segmentation enables capture of individual elasticity behaviors and utilization patterns, directly improving prediction accuracy while maintaining manageable complexity through automated individual analysis

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary analysis of VM resource utilization patterns and elasticity behaviors before overload occurs. By pre-characterizing each VM's behavior patterns and predicting future resource needs, the system can proactively identify potential overload conditions with higher accuracy before they manifest in aggregate metrics

Inventive Principle:

Principle #10Preliminary action

2Productivity

If cloud environments are oversubscribed to operate at high resource efficiency, then resource utilization improves, but the risk of overload conditions increases

Engineering Contradiction:

Improveresource utilizationVSAvoidoverload risk

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent implements feedback mechanisms that continuously monitor actual VM resource utilization against predicted patterns. This feedback loop enables the system to detect deviations from expected behavior and adjust predictions accordingly, allowing reliable overload detection even in oversubscribed environments where traditional methods would fail

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system dynamically adjusts monitoring and prediction parameters based on individual VM behaviors and changing cloud conditions. By adapting analysis parameters to match each VM's specific utilization patterns and elasticity characteristics, the system maintains accurate overload prediction while supporting high resource utilization through oversubscription

Inventive Principle:

Principle #35Parameter changes

3Device complexity

If aggregate history of resource utilization is used for overload forecasting, then data collection is simplified, but the ability to predict future overload deteriorates due to cloud elasticity

Engineering Contradiction:

Improvedata collection complexityVSAvoidoverload prediction accuracy

Core Design Contradiction:

Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments historical data collection to gather and analyze resource utilization metrics for each individual virtual machine rather than collecting only aggregate host-level data. This segmented approach captures the elasticity and utilization patterns of each VM, enabling accurate prediction of future overload conditions while maintaining automated data collection processes

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary characterization of each VM's resource utilization patterns and elasticity behaviors by analyzing historical data individually. This pre-analysis creates baseline models for each VM that enable accurate future predictions without requiring complex real-time aggregate data processing

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11340934B2Cloud oversubscription system

Publication Date: 2022.05.24 AT&T INTELLECTUAL PROPERTY I L P

AI summary

A cloud oversubscription system including one or more processors and a memory coupled with the one or more processors. The one or more processors effectuate operations including obtaining a list of service level agreement (SLA) availability values for each of one or more virtual machines (VMs) of a host. The one or more processors further effectuate operations including analyzing the list to determine a maximum availability number for the host. The one or more processors further effectuate operations including identifying a probable overload condition value based on the SLA availability values. The one or more processors further effectuate operations including performing at least one recommended action when the probable overload condition value exceeds an SLA before an occurrence of an overload condition.