Product Details
Software for measurement and evaluation of performance parameters
Created: 2021
Hruška Martin, Ing. (DITS FIT BUT)
Smrčka Aleš, Ing., Ph.D. (DITS FIT BUT)
anomaly detection,
digital twin,
performance analysis,
performance regressions,
performance testing
Prefekt is an open source tool, that analyses performance profiles captured from analysis and simulation of both digital twins and real traffic of factory production environment (corresponding components, information system, ERP systems, etc.).
The main use case of the Prefekt is for detecting anomalies in performance of individual components in the environment---the machines, the terminals, ERP (Enterprise Resource Planning) systems and the PIS (Production Information System)---as well as during the process of their analysis and simulation.
The tool is mainly meant to be used together with the latest extension of the Tyrant tool, that can generate scenario for simulation of digital twins as well as performance profiles.
The tool (and its process) works in the following phases:
- Prefekt unifies source performance profiles (that can be either in form of a file in JSON format or a query to remote elastic search engine) and creates two data frames (tables) corresponding to baseline (expected) and target (analysed) performance.
- The dataframes are divided to smaller data frames corresponding to each unique type of resources in the profiles (e.g. the duration of the testing, or the CPU usage) and zipped to pairs of baseline and target dataframes.
- For each unique pair of resources types, the correlation coeeficient is computed. Resources that are highly correlated with other resource types (e.g. CPU and memory usage in many use cases) are then skipped from analysis.
- For each pair of baseline and target resources, Prefect detects potential anomalies and reports the difference between the models of the resources as well as effect size of the change.
The code in this project is licensed under BSD-3-Clause.