Quality Attributes Optimization of Software Architecture: Research Challenges and Directions

Quality Attributes Optimization of Software Architecture:
Research Challenges and Directions
Daniele Di Pompeoα Michele Tucciβ
[email protected]α, [email protected]β
α University of L’Aquila, Italy
β Charles University, Czech Republic
MDE4SA @ ICSA 2023, 14 March 2023
Di Pompeo, Tucci Quality Attributes Optimization of Software Architecture: Research Challenges and Directions 1 / 13

View Slide

Motivation
The estimation and improvement of quality attributes in software architectures is a
challenging and time-consuming activity.
Multi-objective optimization approaches used to solve model refactoring optimization
issues12.
Searching among design alternatives (for example, through architectural tactics) is a
typical feature of multi-objective optimization methodologies
1[1] Cortellessa and Di Pompeo. “Analyzing the sensitivity of multi-objective software architecture
refactoring to configuration characteristics”. 2021.
2[2] Ni et al. “Multi-objective software performance optimisation at the architecture level using randomised
search rules”. 2021.

View Slide

A multi-objective model-based framework at a glance
Initial Architecture: The subject
architecture to be optimized
Refactoring actions: The
refactoring actions are the set of all
available actions
selection, mutation, and crossover
to create Architecture Alternatives.
Evaluate fitness function through
fighting objectives. Multi-objective
when there are more than 1 objective.
Many-objective when the objectives
are more than 3
[are the stopping criteria
met?]
Initial Architecture
Architecture Alternatives
Refactoring Actions
Bio-Inspired Actions
Objectives
...

View Slide

Identified Open Challenges
Lack of automation
Problem formalization
Time and resources requirements
Architectural quality metrics
Explainability
Reproducibility

View Slide

Open Challenges: Lack of automation
Search-based techniques require to generate a higher number of alternatives, even
thousands.
Automation to form alternatives is a must-have functionality in every model-based
optimization framework.
Automation is strictly related to the modelling notation and its expressiveness.
Automation would be a step ahead towards the adoption of search-based techniques

View Slide

Open Challenges: Lack of automation
Search-based techniques require to generate a higher number of alternatives, even
thousands.
Automation to form alternatives is a must-have functionality in every model-based
optimization framework.
Automation is strictly related to the modelling notation and its expressiveness.
Automation would be a step ahead towards the adoption of search-based techniques
The first refactoring engine for UML
Our refactoring enginea built atop the Epsilon suite. Employed in a search-based optimization
framework, where we sought optimal solutions with respect to four fighting objectivesb
a[4] Arcelli, Cortellessa, and Di Pompeo. “Performance-driven software model refactoring”. 2018.
b[5] Di Pompeo and Tucci. “Search Budget in Multi-Objective Refactoring Optimization: a Model-Based
Empirical Study”. 2022.

View Slide

Open Challenges: Problem formalization
There not exist guidelines in literature to represent specific problems as chromosomes
Positional structure
Exploited to draw problems in model-based optimization studies [3, 2]. Thus, each position of
the chromosome have a fixed meaning.
+ fastest execution of bio-inspired actions on them.
- too inexpressive within the context of model-based software refactoring.
Action-based structure
+ reports a refactoring action into a chromosome positiona
- slower than the positional structure due to its complexity of compatibility checking
a[4] Arcelli, Cortellessa, and Di Pompeo. “Performance-driven software model refactoring”. 2018.

View Slide

Open Challenges: Problem formalization
There not exist guidelines in literature to represent specific problems as chromosomes
Research Direction
We see an interesting research direction of the problem formalization within model-based
software optimization that should fill the gap with more established search-based optimization
problems.

View Slide

Open Challenges: Time and resources requirements
Improving quality attributes by means of optimization techniques often requires a
considerable amount of time and resources
It is challenging to make model-based SBSE activities more efficient due to the
complexity and the tool-chain employed in these contexts
The process involves a number of transformations from software design models to
non-functional models

View Slide

Open Challenges: Time and resources requirements
Research Direction
This clearly poses a challenge in adopting search-based optimization techniques in
practical software engineering scenarios.
Given the random nature of the algorithms that are usually employed in this context, it is
very difficult to predict how long it will take for the process to complete.
This issue is exacerbated by the fact that, in most cases, the solution space is unknown to
the designer at the beginning of the optimization.

View Slide

Open Challenges: Architectural quality metrics
Quality Indicators only provide numerical feedback for each objective on solutions in the
Pareto front
Compare solutions in a Pareto front with the initial architecture needs a specific indicator
Such a comparison is crucial because it guides the decision-making process of adopting a new
design.

View Slide

Open Challenges: Architectural quality metrics
Research Direction
Quality metrics that represent architectural aspects could make the comparison practical,
and avoid inspecting every solution to obtain enough knowledge to make an informed
decision
We introduce a cost metric to estimate the effort and the distance among models belong
to a Pareto front and the initial onea
a[6] Cortellessa et al. “Many-objective optimization of non-functional attributes based on refactoring of
software models”. 2023.

View Slide

Open Challenges: Explainability
Solutions that are obtained through multi-objective optimization do not carry information
about the specific causes that led to the generation and selection of such solutions.
We cannot explain why some modifications should be applied to our architecture other than
for the quality attributes they seem to improve

View Slide

Open Challenges: Explainability
Solutions that are obtained through multi-objective optimization do not carry information
about the specific causes that led to the generation and selection of such solutions.
We cannot explain why some modifications should be applied to our architecture other than
for the quality attributes they seem to improve
Research Direction
The lack of explainability of results makes it difficult for the designer to justify the new
modifications that she is proposing on the basis of a completely automated process.

View Slide

Open Challenges: Reproducibility
Reproducing results of optimization experiments has been an important concern in recent
years, both for researchers and practitioners.
The solution space is built as the process goes on and the set of architectural elements
that are possible targets of modifications changes.
The main obstacle is represented by the random nature of many optimization techniques
when it comes to the generation of new solutions and the exploration of the solution
space.

View Slide

Open Challenges: Reproducibility
Reproducing results of optimization experiments has been an important concern in recent
years, both for researchers and practitioners.
The solution space is built as the process goes on and the set of architectural elements
that are possible targets of modifications changes.
The main obstacle is represented by the random nature of many optimization techniques
when it comes to the generation of new solutions and the exploration of the solution
space.
Research Direction
Achieving perfect reproducibility is still a challenge in architectural optimization, and one that
is rarely addressed by the relevant literature.

View Slide

Future Work
Our agenda foster new research activities in the field of software architecture optimization,
especially for the optimization of non-functional properties, such as performance and reliability.
As short-term future work, we will try to tackle the challenge about the lack of
automation3
As mid-term future work, we will attempt to address the challenge about the architectural
quality metrics by exploiting quality estimation techniques well-recognized in the
search-based community4
As long-term future work, we will attempt to address the challenge about the problem
formalization with an empirical analysis of several software architectures
3[7] Cortellessa et al. “On the impact of Performance Antipatterns in multi-objective software model
refactoring optimization”. 2021; [8] Arcelli et al. “EASIER: an Evolutionary Approach for multi-objective
Software archItecturE Refactoring”. 2018.
4[10] Li, Chen, and Yao. “How to Evaluate Solutions in Pareto-Based Search-Based Software Engineering: A
Critical Review and Methodological Guidance”. 2022.

View Slide

Conclusion
We reported open challenges of quality attributes optimization of software architectures.
These open challenges hinder the utilization of search-based techniques in software
architecture optimization.

View Slide

Any Questions?
Quality Attributes Optimization of Software Architecture:
Research Challenges and Directions
Daniele Di Pompeoα Michele Tucciβ
[email protected]α, [email protected]β
α University of L’Aquila, Italy
β Charles University, Czech Republic
MDE4SA @ ICSA 2023, 14 March 2023

View Slide

Quality Attributes Optimization of Software Architecture: Research Challenges and Directions

Quality Attributes Optimization of Software Architecture: Research Challenges and Directions

Daniele Di Pompeo

More Decks by Daniele Di Pompeo

Other Decks in Research

Featured

Transcript