IT Performance Modeling via Clustering and Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity and rapid evolution of IT systems make traditional performance modeling expensive, time-consuming, and intrusive, especially in non-stationary production environments with volatile workloads and limited monitoring capabilities, necessitating a more efficient and automated approach for online performance assessment and capacity planning.
Innovation Solution
The method employs clustering techniques to reorganize performance measurement data into multiple regimes, using advanced inference technologies to build queuing network models automatically, incorporating end-to-end response times and server utilization, and applying optimization algorithms to infer service demand parameters, enabling fast and accurate performance modeling and optimization without extensive manual tuning or intrusive monitoring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional performance modeling methods are used on large complex IT systems, then modeling accuracy can be maintained for small systems, but the cost and time required become prohibitively expensive and time-consuming
Solution Approach 1:
The patent segments the complex IT system into multiple hierarchical levels (infrastructure level, platform level, application level) and models each level separately using appropriate queuing network models. This allows accurate modeling of large systems by breaking them down into manageable components that can be modeled and analyzed independently while maintaining overall system accuracy.
Solution Approach 2:
The patent creates simplified representative models (copies) of the actual IT system at different abstraction levels. These models replicate the essential performance characteristics and relationships without requiring detailed modeling of every component, enabling fast and accurate performance assessment of large complex systems.
2Measurement precision
If detailed modeling parameters are collected through direct measurement, then model accuracy improves, but the monitoring becomes intrusive and costly
Solution Approach 1:
The patent introduces performance counters and monitoring agents as intermediaries between the production system and the modeling process. These intermediaries collect necessary performance data (CPU utilization, memory usage, transaction rates) without directly interfering with system operations, and provide the data needed for accurate queuing network model parameter estimation.
Solution Approach 2:
The patent replaces intrusive direct measurement methods with indirect inference through queuing network models. Instead of measuring detailed internal parameters directly, the system uses easily obtainable performance metrics and mathematical models to infer the necessary modeling parameters, reducing intrusiveness while maintaining accuracy.
3Measurement precision
If performance modeling is done manually with detailed tuning, then model accuracy can be achieved, but labor costs and complexity increase significantly
Solution Approach 1:
The patent implements automated model building and tuning capabilities where the system itself performs the modeling process. The queuing network models are automatically constructed from system architecture information, and parameters are automatically estimated from performance data using statistical methods, eliminating the need for manual model tuning and reducing both labor costs and complexity.
Solution Approach 2:
The patent uses parameter estimation techniques that automatically adjust model parameters based on observed performance data. The system changes parameters iteratively to match measured performance metrics, achieving accurate models without manual intervention in the complex parameter tuning process.
4Measurement precision
If extensive monitoring infrastructure is deployed to collect performance data, then data quality improves, but equipment costs and system complexity increase
Solution Approach 1:
The patent designs the monitoring infrastructure to serve multiple purposes: collecting performance data for model parameter estimation, validating model predictions, and providing real-time system performance monitoring. This multi-functionality reduces the need for dedicated monitoring equipment while improving data quality through broader data collection.
Data Source
AI summary
A system and method for performance modeling for an information technology (IT) system having a server(s) for performing a number of types of transactions includes receiving data for system topology and transaction flows and receiving performance measurement data for the IT system. The measurement data is clustered into multiple regimes based on similarities. Service demand and network delay parameters may be inferred based on clustered data.