Software Evolution of Product Families in High-tech Equipment
report
Variability and evolution are key drivers of complexity in high-tech equipment. Every product has many variation points, resulting in a large number of unique system configurations. These systems must be maintained throughout their life-time, which may last several decades. During this time, digital technologies, e.g. containerization and orchestration technologies, may become deprecated or evolve many times. Updating configurations and implementations for each system variant to account for such changes is both costly and time-consuming. To make matters worse, changing software technologies affects technical
system performance, requiring technical performance to be re-optimized and re-verified. The TechFlex project is a collaboration between TNO-ESI and Thales that addresses the challenge of variability and evolution in software-intensive high-tech equipment. This is done by investigating to what extent it is possible to specify the software configuration in a technology-agnostic way yet automatically generate efficient software deployments that satisfy performance requirements. The result of this research is a model-based methodology to specification and automation with two steps.
1. Technology-agnostic specification of software configurations based on a family of domain-specific languages (DSLs) with technology-specific generators that produce artifacts to create a custom software deployment with minimum manual intervention.
2. Deployment optimization based on model-based reinforcement learning using Monte Carlo Tree Search (MCTS) that improves the mapping to software processes to compute nodes to ensure technical performance requirements are satisfied.
This report describes the methodology in context of a fictive Meal Delivery System (MOS), a simple case study inspired by an application in the defense domain. An evaluation of the approach shows that the technology-agnostic DSLs are expressive enough to describe the configuration of the MOS, as well as a confidential industrial case study.
system performance, requiring technical performance to be re-optimized and re-verified. The TechFlex project is a collaboration between TNO-ESI and Thales that addresses the challenge of variability and evolution in software-intensive high-tech equipment. This is done by investigating to what extent it is possible to specify the software configuration in a technology-agnostic way yet automatically generate efficient software deployments that satisfy performance requirements. The result of this research is a model-based methodology to specification and automation with two steps.
1. Technology-agnostic specification of software configurations based on a family of domain-specific languages (DSLs) with technology-specific generators that produce artifacts to create a custom software deployment with minimum manual intervention.
2. Deployment optimization based on model-based reinforcement learning using Monte Carlo Tree Search (MCTS) that improves the mapping to software processes to compute nodes to ensure technical performance requirements are satisfied.
This report describes the methodology in context of a fictive Meal Delivery System (MOS), a simple case study inspired by an application in the defense domain. An evaluation of the approach shows that the technology-agnostic DSLs are expressive enough to describe the configuration of the MOS, as well as a confidential industrial case study.
Topics
TNO Identifier
1013704
Publisher
TNO
Collation
29 p.