Migration from single to multicore ECU systems are always challenging. One such challenge is ensuring consistent state of shared data among several program components executing on different cores. We have developed an eclipse based eco system to support developers for analyzing such use cases based on Model Driven System Analysis (MDSA). This involves eclipse technologies such as APP4MC, EMF and SIRIUS. This framework combines architecture description with generated code to prepare a base for data consistency analysis.
In this talk, we would like to share our experience of development and deployment challenges of this Data Consistency Checks (DCC) framework. This framework has several aspects, from model preparation to report assessment.
Following aspects of the framework will be highlighted:
- Model generation from source code (C code and ELF)
- Definition of use case specific data consistency criterion/constrains
- Error reporting and trend analysis
- Visualization of complete error hierarchy
- Over vs. Under synchronization
- Scalability as in handling of largescale data set
Further, we shall share our experience of using this framework in the largescale integrated multi core industrial systems, which consist of TIER-1 and OEM code. With DCC, we are able to substantially increase the accuracy of reported errors by reducing the number of false positives through general approaches and at the same time processing users input allowing series of domain-specific patterns.
We would also like to share the future trend of DCC, which already goes from "know your customer" to "learn from past executions (of DCC)". Further unwanted false positives we can reduce by introducing machine-learning concepts for enhanced data analysis on huge data sets. Allowing supervised learning strategies by taking user feedback in an automized way can be one of many existing approaches we are going to share with you.