Reliability Estimation for Ad Hoc Applications via Dependency Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for measuring reliability in distributed systems with disparate components and obscured infrastructure are inadequate, as they require access to proprietary network schematics and assume independent failures, leading to coarse-grain reliability estimates.
Innovation Solution
A system that uses user-input application definitions and relationships between computing resources to generate conditional probability tables, estimating reliability through successive trials based on historical availability data, allowing for accurate reliability estimation without disclosing infrastructure details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If formal models and end-to-end system descriptions are used to measure reliability, then reliability measurement is possible, but the measurement is inadequate in distributed systems with obscured infrastructure and disparate components
Solution Approach 1:
The patent introduces an intermediary layer (the reliability estimation system) that sits between the user and the obscured infrastructure. This intermediary uses probabilistic models and historical data to translate unavailable detailed infrastructure information into meaningful reliability estimates, allowing measurement without direct access to internal system details.
Solution Approach 2:
The patent changes the parameters used for reliability measurement from requiring detailed infrastructure specifications to using observable inputs (application definitions, component relationships) and historical availability data. This parameter transformation enables reliability estimation in systems where traditional detailed models are unavailable.
2Reliability
If proprietary network schematics are required for reliability measurement, then accurate reliability data can be obtained, but infrastructure details must be disclosed
Solution Approach 1:
The patent extracts only the essential information needed for reliability estimation (application definitions, component relationships, historical availability) while leaving the proprietary infrastructure details (network schematics, internal architecture) behind. This selective extraction provides accurate reliability data without requiring disclosure of sensitive infrastructure information.
Solution Approach 2:
Instead of requiring access to the actual proprietary infrastructure, the system creates a probabilistic model (a copy) that captures the essential reliability characteristics. This model copy allows reliability analysis without exposing the real infrastructure details, maintaining security while enabling measurement.
3Device complexity
If simple uncorrelated failure modes are assumed, then reliability calculation is simplified, but the estimation becomes coarse-grain and inadequate for complicated systems
Solution Approach 1:
The patent introduces dynamic probabilistic models that can adapt to the specific characteristics of complicated distributed systems. Rather than using static simplified failure assumptions, the system uses dynamic probabilistic relationships that capture correlations and dependencies among components, achieving both tractability and precision in complex environments.
Data Source
AI summary
In certain embodiments, a computer-implemented method includes receiving a request for a reliability estimate associated with an ad hoc application. In response to the request, one or more components associated with the ad hoc application and upon which the ad hoc application relies are identified. The method also includes generating a directed graph. The directed graph identifies one or more dependency relationships among the identified components. The method also includes calculating, based at least in part on the directed graph, a reliability estimate for the ad hoc application.