Performance Engineering in High-Tech Systems: A Visit to CISTER Research Center

I recently had the opportunity to visit the CISTER Research Center for Real-time and Embedded Computing in Porto. It was a great chance to reconnect with former colleagues from my previous tenure there. During my visit, I was invited give a presentation focusing on the complexity challenges in high-tech equipment and ESI’s vision on model-driven performance engineering in this field. The talk also highlighted the growing use of microservice architectures in cyber-physical systems to address the complexity drivers. I explained that while there are good open source tools for instrumenting applications and gathering telemetry data, such as metrics, logs, and traces, new automated analysis methods are needed to reduce the effort of optimizing, verifying, and diagnosing system performance. In this context, I introduced a framework for telemetry-based performance engineering that can be used to address a number of performance challenges. In particular, I explained how it could be used to check whether the system implementation conforms to a UML specification, both in terms of timing and behavior, and to perform performance prediction.

The presentation, which was attended by a group of approximately 15 staff members and PhD students, was well-received and led to fruitful discussions about the relation between real-time systems research and performance engineering. . The visit concluded wonderfully with a delightful dinner with the institute’s director, Prof. Eduardo Tovar, in Matosinhos.

Automatic Workload Inference Improves Scalability of DSE in Complex Systems

I am happy to announce that the paper “Automated Derivation of Application Workload Models for Design Space Exploration of Industrial Distributed Cyber-Physical Systems” has been accepted for publication at the 7th IEEE International Conference on Industrial Cyber-Physical Systems (ICPS). The paper is first-authored by Faezeh Saadatmand in the context of the DSE2.0 project, a part of the academic research program MasCot, co-funded by TNO-ESI and NWO. Congratulations Faezeh!

The paper addresses challenges with respect to designing their next-generation distributed cyber-physical systems (dCPS). Efficient Design Space Exploration (DSE) techniques are needed to evaluate possible design decisions and their consequences on non-functional aspects of the systems. To enable scalable and efficient DSE of complex dCPS, it is essential to have abstract and coarse-grained models that are both accurate and capable of capturing dynamic application workloads. However, manually creating such models is time-consuming and error-prone, and they need to be continuously updated as the system evolves. This research addresses this need by introducing an automatic method for deriving an application workload model. This model, based on trace analysis, captures computation and communication activities within an application in a timing-agnostic manner. The approach has been validated through a case study on an ASML Twinscan lithography machine, demonstrating high accuracy in capturing real application workloads. Next steps in this research involves combining this model with an automatically inferred hardware platform model to enable DSE exploring different hardware, software, and mapping alternatives.