Application Deployment Reliability Assessment in Data Centers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deploying applications in large-scale data centers faces challenges due to the NP-hard nature of finding reliable deployments, where redundancy efforts can be undermined by shared hardware or software dependencies, limiting developers' control over placement and making it difficult to assess and ensure application reliability.
Innovation Solution
A computer-implemented method assesses the reliability of application deployments by determining failure states for components, evaluating reachability, and selecting deployments based on comparison, using Monte-Carlo sampling and variance reduction techniques to identify a deployment plan that meets reliability criteria, even in complex topologies like Fat Tree, BCube, or DCell.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple replicas of an application are deployed to increase reliability, then the reliability of the application is improved, but the reliability can be undermined by common hardware or software dependencies shared by replicas
Solution Approach 1:
The patent segments the deployment assessment into multiple independent reliability determinations for different deployment configurations. Each deployment plan is evaluated separately by determining reliability information through component failure state analysis, allowing complex deployment scenarios to be broken down into manageable assessment units that can be compared independently
Solution Approach 2:
The patent performs preliminary reliability assessment and comparison of multiple deployment plans before actual application deployment. By determining reliability information for each predetermined deployment in advance and comparing them, the system identifies the most reliable deployment configuration beforehand, preventing reliability issues before they occur
2Reliability
If developers have limited control over placement of application in a data center, then deployment flexibility is reduced, but finding a reliable deployment becomes NP-hard
Solution Approach 1:
The patent introduces an intermediary assessment mechanism that evaluates deployment reliability without requiring direct control over data center placement. The system acts as a mediator by analyzing component failure states and reachability information to determine reliability metrics, enabling developers to assess reliability even when placement control is limited
Solution Approach 2:
The patent changes the assessment parameters from direct placement control to reliability metric evaluation. By determining reliability information based on component failure states and reachability rather than direct placement decisions, the system transforms the NP-hard placement problem into a more manageable reliability assessment problem with comparable metrics
3Reliability
If redundancy is used to increase reliability, then application availability is improved, but common hardware or software dependencies can cause simultaneous failure of replicas
Solution Approach 1:
The patent applies local quality analysis by evaluating the reliability of individual components and their failure states within the broader deployment system. By determining reachability information for each component and analyzing their specific failure modes, the system identifies local dependency vulnerabilities that could cause simultaneous replica failures, allowing targeted mitigation of common dependency risks
Data Source
Figure 1
Figure 2
AI summary
The invention concerns a Method and corresponding device of deploying an application reliably in a computer network (100) comprising: Determining (204) information about a reliability of a predetermined first deployment for the application, Determining (207) information about a reliability of a predetermined second deployment for the application, Comparing (208) the information about the reliability of the first deployment with the information about the reliability of the second deployment, Selecting (208) the first deployment or the second deployment for deploying the application in the computer network (100) depending on the result of the comparison.