SE

<< back to topics

Variability & Software Product Lines

Summary of Selected Papers

Most products, such as cars, printers, mobile phones, etc., exist in various variants. Software for product variants is quite similar, but typically differs in new or additional features that sometimes deeply affect the software's architecture. Software variants are managed as a Software Product Lines (SPLs) that captures the commonalities as well as the differences. Software Product Lines have many benefits, they:

  • decrease development time of new product variants,
  • decrease time to market,
  • lead to better software quality,
  • improve reuse, and
  • reduce bug fix time.

Variability is to a larger extent related to evolution. We discuss our approaches to evolution understanding in evolution & transformation of models.

Feature diagrams and Views

Feature diagrams describe variability in a top down fashion in the problem space. We studied the application of this top down approach, e.g., in the automotive domain in [GKPR08]. They suffer from the need to first decompose the problem space and understand possible features in order to build the feature diagram before being able to apply it. In [GHK+08] and [GKPR08] we also speak of a 150% model. This normally enforces a product line definition phase in which the requirements and features need to be collected which creates additional costs. Among others we discuss decreasing these costs in [GRJA12].

Delta Modeling

We discuss delta modeling as a bottom up SPL modeling technique in [HRR+11]. Deltas can both be used as subsitute and as extension to traditional feature based development. Deltas allow us to build a product line incrementally starting with a base variant when the need for a new feature arises. Starting with a core version, each delta describes the changes necessary to derive a new variant. Deltas allow to add, replace, modify and delete components of a model resp. implementation and is thus rather general.

Each set of valid deltas configures a product variant. We have successfully applied delta modeling to the architectural analysis and design language (ADL) MontiArc by creating Delta-MontiArc ([HRR+11] and [HRRS12]) as well as applied it to Simulink creating Delta-Simulink [HKM+13]. Deltas can not only describe spacial variability but also temporal variability which allows using them for software product line evolution [HRRS12].

In [HHK+15] we have generalized this approach to the general question, how to synthesize a delta modeling language based on a given modeling language. Thus deltas can generally be applied to other languages too.

Variability in Language Definitions

On a related line of research, we also have studied variability of modeling languages, which allows us to define and reason about syntactic and semantic variation points, which is, e.g., in the UML a big topic as it seems the UML standard will otherwise not be able to accommodate all stakeholder requirements.

For this purpose we defined a systematic way to define variants of modeling languages [CGR09]. We applied this research, e.g., in the form of semantic language refinement on state charts in [GR11]. Additionally, we investigated a method to model syntactic language variability through language product lines .

In [FPR02] we discussed how to apply annotation to the UML to describe product variation points.

SPL and Delta Modeling in Industry

We have introduced SPL and delta modeling in several companies and are proud of succesfully helping companies to manage their variants. We also learnt, that industrial success means that each company needs a tailored process that fits the company culture, used tool chains, size of products and the desired agility of variant construction. SPL does not come free of initial cost.

A typical SPL introduction process consists of three stages: (1) Understanding the current situation in the company. Current process? Size of projects? Number of existing and planned variants? How similar are those? Current costs of evolution for individual products? Available and desired tool chain? (2) Derivation of a long list of potential technical, process and organizational measures for an SPL based future with efficient development of high quality systems. Categorization and priorization. (3) Implementing the most promising steps and understand the effects.

Summary

  1. Products often exist in various variants. Variants are managed in a Software Product Line (SPL) that captures commonalities and differences.
  2. Modeling variants explicitly is essential to tame complexity of otherwise 150% models.
  3. Delta modeling is a bottom-up technique starting with a running base variant and can more easily be applied to derive and implement SPLs from existing products in industrial size projects.
  4. Delta-concepts are successfully applied to MontiArc and Simulink.
  5. Variability applied to modelling languages allows to capture semantic variation. This is essential to capture the semantics of the UML.
  6. Industrial introduction of SPL based development needs organizational and process adaptation and needs to be tailored to the individual company.

Related Topics

Selected Related Publications



<< list of all topics list of all publications >>