Model-based software development – its real benefit
1. Introduction
Model-based development becomes more and more popular in the development of embedded software systems in the car industry. On the websites of tool vendors many success stories can be found, which report efficiency gains from up to 50% in the development, high error reductions and a more rapid increase of the maturity level of developed functions just because of model-based development. In the automotive industry there is a controversy about the benefit of model-based software development.
There are rumors in the industry that some companies report about cost savings and quality improvements because of model-based development and some companies report of no major changes or even about cost increases. This uncertainty, along with high upfront costs for the introduction of model-based development, even holds companies off from switching. The companies who have switched are interested how they can optimize the economics of their model-based development process, especially when they have no major benefits with a model-based development approach. Reliable and broadly spread research that analyses the status quo of model-based development and its effects on the economics are still missing. This article describes the results of a global study by Altran Technologies, the chair of software and systems engineering and the chair of Information Management of the University of Technology in Munich which examines the costs and benefits of model-based development of embedded systems in the car industry.
We want to find out what effects model-based development has on cost-, time- and quality changes and in addition identify factors which do have an important influence on the costs. These factors can be used by the companies as set screws to optimize the economics of their model-based development process.
In the following article a selection of the results of the study is described. At the beginning we focus on the development phases in which model-based development is actually being used, present the status of the seamlessness in the development process and describe in how far the study participants expectations in model-based development have been fullfield. Afterwards we analyze the effects on the economics: First of all generally in a total view and following detailed using the example of requirements engineering. Statements to further questions like for example cost- and quality improvement potentials from the study participants view and recommendations for action on the basis of the study results can be found in the full results of the study.
2. Status quo of the usage of model-based development and the status quo of the seamlessness in the development process
The study shows that model-based development is being used for series development by the majority of the companies for more than five years. Especially in the development phases software design and implementation model-based design is used intensively. 96% of all participants use model-based development in both development phases. The other 4% have just begun using model-based development and use it in the software design. 75% of the companies use model-based design already in the requirements engineering phase by using Rapid Control Prototyping (RCP). We want to keep in mind that RCP is not model-based requirements engineering. Instead it is already designing with the modeling tool. Nevertheless the study participants report that it is an efficient method to identify missing requirements, especially while developing innovative functions. A trend can be seen that more and more companies use model-based design also in the architecture development. The aim is to increase the seamlessness in the development process and to perform early tests on the architecture model. Model-based testing (i.e. the generation of test cases out of a test model) is actually not used intensively. Only 35% of the participants use it right now, but almost 50% plan to use it in the near future.
After describing the status quo of the model-based design activities in the car industry we like to present the status quo of the seamlessness in the development process, an important factor when it comes to improve the efficiency in a model-based development. 96% of all study participants report of a seamlessness between the development phases software design and implementation. The seamlessness is ensured by using a code generator which transforms the function model in C-Code.
However between the other development phases there is still lots of room to improve the seamlessness. 28% of the participants report of a seamlessness between requirements engineering and the architecture design and 46% between the architecture and the software design. With the correlation of the seamlessness in the development and the know-how of the employees it could be discovered that the know-how of the employees has a huge influence on the seamlessness. The higher the know-how of the employees is the higher is the reported seamlessness in the development process. This is especially the case for the seamlessness between requirements engineering and architecture design and architecture design and software design. Companies with employees with high know-how report more than twice as much of a seamlessness in these development phases than companies where the employees have little know-how.
The study participants report in general of fulfilled expectations when it comes to judging the fulfillment of the expectation they had when deciding to switch to a model-based approach. In the correlation with the reported cost savings it can be seen that the fullfield expectations depend significantly on the height of the cost savings. The higher the savings are the better are the statements to the fullfield expectations. However the majority of the participants report that there is still plenty of room for improvement to exploit the full potential of a model-based approach.
3. Effects of model-based development on the economics
The main aspect of the research was the economics of a model-based development. In the following central results are being outlined. Detailed information, especially the effect of model-based development in each single development phase, can be found in the full results of the study.
The companies report in average of cost savings around -27% and time savings around -36%. Figure 1 shows the relative effects of a model-based design in each development phase and also in a total view. In the development phases requirements engineering, architecture design and software design considerable cost increases can be seen. They result of the frontloading of the development activities like performing Rapid Control Prototyping (RCP) already in the requirements engineering, modeling the architecture in formal models and the modeling of the complete logic in a function model with adjacent respectively accompanying model tests (e.g. MiL-Tests, model reviews, RCP and guideline checkers). In the implementation and in the test phase the companies report of cost savings, because they use the possibility of automatic code generation and benefit significantly of the early error detection (e.g. in the software design) and the thereby associated savings from iterations in the development.
Figure 1: Relative cost changes in the single development phases
The data analysis exposed that the statements of the study participants concerning the changes in the total cost partially deviate considerably. Some study participants report of considerable cost savings whereas others report of considerable cost increases. We found out, that study participants with less than one year experience generally report of cost increases. These result on the one hand of a not well-rehearsed development process and on the other hand of little know-how of the employees in the use of the development tools. Companies with more than one year experience report in general of cost savings. However also in this group there are huge differences in the statements about the saving amount. The data analysis showed the top three factors which influence the costs most are:
– A high modeling degree of the function model and consequently a high degree of generated code
– Intensive test activities on function model level
– High know-how of the employees in model-based design as well as in software engineering
During the data analysis we found out that study participants who have high values in all three criteria have reported of the highest cost savings which were in the area of around 40%. As these companies had all long experience in model-based development it is important to notice that an established development process with a high seamlessness is a prerequisite for cost savings compared to a hand coded development.
In the following we like to show the influence of model-based development in the example of requirements engineering.
In requirements engineering RCP plays a major role, especially in the development of innovative functions. We previously pointed out that RCP means to already model with the modeling tool for the software design. Nevertheless we like to present the effects of RCP as this also shows which potential lies in “modeling” the requirements.
First of all we like to present the effect on the development costs in the requirements engineering phase. Because of the frontloading activities the costs raise with increasing modeling degree of the RCP-model. In average the study participants, who use RCP, report of a cost increase of around 23%.
Figure2: Correlation modeling degree RCP and changes in costs requirements engineering
After analyzing the effects on the costs it is interesting to see what effect the conduction of RCP-tests has on the number of detected errors in requirements engineering. Figure 3 shows impressing that with a rising modeling degree in the RCP model the number of detected errors in the requirements engineering phase increases.
Figure 3: Correlation modeling degree vs. number of detected errors
Besides the number of detected errors we also analyzed the effect of RCP on quality criteria for requirements engineering defined by IEEE98. Figure 4 shows the results, where –means strong decline, – decline, 0 no changes, + increase and ++ strong increase.
Figure 4: Changes in the quality criteria in the requirements engineering when using RCP
It can be seen that the study participants report of improvements in all quality criteria. Especially in the criteria completeness, correctness and verifiability major improvements have been reported.
When correlating the modeling degree in the RCP-model with the changes in the total development costs, it can be seen that companies that use RCP have higher savings in total than companies that don´t. A clear trend like the higher the modeling degree in the RCP-model the higher are the total cost savings can´t be reported. The results recommend a modeling degree of the RCP-model less than 60%.
Figure 5: Correlation modeling degree RCP vs. changes in the total costs
4. Summary
In this article we asked the question if model-based design leads to cost savings in the development. The answer to this question is yes. The results of the study show that with a right approach model-based design can bring significant cost savings and also improvements in the product quality.
However a model-based design doesn´t guarantee to be better than a hand coded process. If model-based design is not used the “right” way (e.g. only punctual model-based development) or the development process is not established (i.e. process interfaces and roles are not clear or the employees are still not used to the new development approach and the new development tools) development cost increases can result. This is especially the case right after the process redesign. There is a detailed know-how necessary to establish a model-based development process in a company and to optimize this process to be more efficient in the development e.g. by raising the seamlessness of the development tools with own developed tools (e.g. scripts). The companies can establish this know-how slowly on their own or much faster with the support of a technology consulting company, which has experience for years in the field of model-based design.
About the authors:
Prof. Dr. Dr. h.c. Manfred Broy: Owner of the chair Software & Systems Engineering of the technical university in Munich www4.in.tum.de
Prof. Dr. Helmut Krcmar: Owner of the chair Information Management of the technical university in Munich www.winfobase.de
Dr. Jens Zimmermann: Solution manager Software Altran Technologies and responsible person for the study www.altran.de
Sascha Kirstan: Engineering Consultant at Altran Technologies and external PhD-student at the chair software and systems engineering of the technical university in Munich.
Responsible for the conduction of the study and the evaluation of the data: www.altran.de
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
