[Context and Motivation:] Complex, dynamic, and distributed systems confront requirements engineers with new challenges. The systems operate in dynamic environments with an increasing complexity. Modeling the context for understanding, developing and validate the requirements becomes a di cult task. [Question/ problem:] Current requirements engineering techniques for gathering information are not su cient to capture the system behavior and the environment for this class of systems. [Principal ideas/ results:] A new approach is needed to support the development of an environment model, the idea is to use runtime information (e.g., from executing early prototypes of the system) to gather additional facts about the environment. [Contribution:] In this paper I present the motivation and challenges for the environment modeling for complex, dynamic, and distributed systems and an overview of the progress of my thesis. For this purpose I present my previous work on this eld and discuss the proposed work for my thesis.
The part of the real world that is relevant to understand a system is usually called its (operational) environment (we consider this term as synonymous to context in this paper). A proper understanding of the environment is a prerequisite to write requirements for a system.
There are characteristics of a system that complicate the gathering of information on the environment and its requirements. These characteristics are: { Dynamic: the behavior of a dynamic system is a ected by changing users' needs and uncertainties in their operational environment. These systems often have a considerable lifetime and operate in di erent situations, which leads to evolving or changing requirements. A system needs feedback mechanisms to analyze the di erent environment situations and adapt to satisfy the requirements. { Distribution: a distributed system consists of components located on networked computers, which communicate and coordinate their actions. They share di erent resources and capabilities to reach a common goal or collaborate to reach individual goals.
Examples for these characteristics can be found in the mobility area. A future car for instance will be connected with other cars or with parts of the tra c infrastructure, like intelligent tra c lights. Theses connected entities leads to a dynamic environment for the car with services and information that are not always available. If we look at autonomous driving, the complexity of the environment and the resulting requirements is immense. The mobility domain is just one example for systems with these characteristics. Other examples can be found in the Internet of Things (IoT) domain.
Problem. The classic RE techniques for gathering information are not fully su cient to capture the environment and the requirements of such a system characterized above. One reason is that the information about environment and system behavior available from stakeholders is a priori incomplete. The missing information leads to incomplete speci cations and incorrect software.
Due to the dynamic nature of such systems, requirements evolve during runtime and are a priori not completely available or not fully understood to elicit them with techniques like interviews.
Newer approaches to requirements engineering, e.g., from the eld of
selfadaptive systems [
Goal. The goal of my thesis is to provide a better understanding of environment modeling for complex, dynamic and distributed systems. This includes (1) the identi cation of missing concepts of environment models and (2) the adoption of techniques to collect this information. The idea is to adopt techniques beyond the RE phase, more precisely from downstream software development: e.g., prototyping, simulation, and eld test. An implementation of a RE tool to support the improved modeling process will be provided as a proof of concept.
The proposed work will address the following research questions: [RQ1] What are the core concepts for dynamic and distributed systems to become represented in environment models and how to represent them? [RQ2] What methods are useful to identify the relevant information for environment models beyond the RE phase? [RQ3] What kind of information are needed by an requirements engineer to improve the development process and the adaption process? Su cient information about the environment of a system are needed to develop, understand, and validate requirements. An adaptive system uses the environment information for a better decision space and to improve the overall adaption process.
Solution. Our rst approach towards this goal was to weave runtime
information about the requirements and variable parameters of a dynamic system back
into the requirements speci cation, see [
Based on this approach we consider an iterative, tool-supported process for gaining runtime information about the environment and the requirements, which is depicted in Figure 1. First, the requirements are gathered and an environment model will be modeled by a requirements engineer. Then, a prototyping and evaluation framework is used to support the development of a system. The frameworks can be used to generate probes for an application and the whole adaption mechanism (with integrated MAPE loop) out of a speci cation. It includes an instrumentation for simulation and eld test, which collects runtime data on environmental entities. We are convinced that an early eld test can help to nd missing environment data and to improve the general knowledge about the system behavior. Simulations can be used as an alternative to the eld test, if the environment model contains su cient information.
Based on the information gathered during a eld test or simulation - the requirements and the environment model enhanced in the next iteration. The before-mentioned information is gathered by tracing components inside of the prototype and by observations and feedback from the requirements engineer.
In my thesis I focus on the environment model as it is the most important artifact in the process. It is used for the generation of the prototype, to describe the environment for simulation, and to collect the environment information gathered during simulation and early eld test.
The remainder of the paper is structured as follows. Sec. 2 represent a short summary of a systematic literature review on environment modeling. Sec. 3 describes the process of my PhD thesis with an overview on previous work, Sec. 3.1, and a discussion on the proposed work, Sec. 3.2. 2
We performed a systematic literature review (SLR) on environment modeling for adaptive systems. The SLR provides rst results for the research question [RQ1] und [RQ2]. The search strings are based on three main facets: research object environment model, quali cation - adaptive and restriction - requirements engineering and the alternative terms. We have chosen IEEE Xplore Digital Library, Springer Link and Google Scholar as databases for identifying the publications. At the start of the selection process we gathered 455 publications, which were tested by prede ned selection and exclusion criteria. This process resulted in 58 relevant publications.
In the SLR, we considered elicitation and validation of the environment. Unexpected for us are the few results for these activities. Elicitation is addressed mostly with a few sentences that refer to interviews and meetings. Only one paper compares di erent elicitation methods like brainstorming, 6-3-5 method, six thinking hats, eld observation, apprenticing and interviews. We expected more focus on eld observation, because the information about the environment is a priori often not available as mentioned previously.
In conclusion, we argue that we need a systematic process for eliciting, modeling, and validating the environment: an environmental model provides a reference for requirements, it provides a basis to anticipate changes in the environment, it allows to derive test scenarios, and it helps to monitor the environment at runtime.
We identi ed four major modeling approaches: goal-based, agent-based, rulebased, and state machine modeling. As an example for an approach to model the environment we discuss the goal-oriented approach. A goal-oriented approach focuses on modeling the goals of an actor and how a goal can be achieved over alternative ways. The modeling process trying to nd actors and models the relationship between them. There are no opportunities to model a system boundary. It is not relevant for the modeling approach if an actor is part of the system or not, it is important what parts belong to a speci c actor and how the actors depend on each other. Additional elements or annotations are used to identify context elements.
Figure 2 shows two di erent ways to represent the environment (i* Example).
First, it is a physical or virtual element in the environment of an actor, which is
needed to achieve a goal of this actor. The element is represented as a resource
element [
Second, the environment is represented as actor itself [
The starting point of the thesis is the iterative enhancement process (shown in Figure 1) tailored for complex, dynamic, and distributed systems. Two empirical studies are planned to analyze and compare the identi ed approaches for environment modeling. Based on the results a new integrating environment model will be developed and integrated in a RE tool. The tool and the process will be analyze and validated with empirical studies.
In the remainder of this section the previous performed work will be presented and the plans for the rest of the thesis will be discussed. 3.1
As mentioned previously, our rst approach in this area was a feedback-aware
requirements document in [
Next we suggested a prototyping and evaluation framework for
selfadaptive systems in [
For bench-marking we propose a case from the IoT domain, smart cities
in particular, which comprises of hardware and software components, see [
A systematic literature review on environment modeling for adaptive systems was performed, in particular from a requirements perspective. We addressed the goals, the modeling concepts as well as the methodological aspects of environment modeling in our survey. As major results of our survey, we provide an ontology of typical goals of environment models in adaptive system research and a meta-model of existing environment modeling concepts. As a negative nding { and a research opportunity { we nd that so far methodological aspects of environmental modeling for requirements engineering have received very little attention. 3.2
Currently, I'm working on a more speci c conception of the environment. This includes my view on relevant and irrelevant parts or information that should be included in an environment model - given the extended ways of using it.
I am planing to enlarge the case study on IoT as a base for empirical studies to evaluate the identi ed approaches for environment modeling from our survey. The advantages and disadvantages will be determined. Based on the gathered information a general meta-model for environment modeling will be developed.
As a result of my PhD thesis, I will develop a prototype RE tool based on the meta model that includes supporting mechanisms to create environmental models and to enrich this model by prototyping, simulation, and eld test.
The resulting RE tool will be validated using the IoT case mentioned above and if possible at that time in cooperation with a local company. 4
This research was supported by graduate school of University of Applied Sciences and Arts, Dortmund.