CSMR11b.ppt
1 like313 views
This document summarizes a research paper on identifying micro-architectures in evolving object-oriented software systems. The paper presents an approach called SGFinder that models class diagrams as labeled graphs and defines micro-architectures as connected induced subgraphs. SGFinder efficiently enumerates all micro-architectures up to a given size. The paper applies SGFinder to two open-source systems and analyzes the identified micro-architectures to find those that are particularly fault-prone, fault-free, stable or change-prone. The results provide insights into common micro-architecture patterns and their relationships to quality attributes.
![Related work (I)
In general, the proposed approaches rely on a library of known motifs
and use architectural recovery techniques based on sub-graph matching
or on motifs properties.
Constraint Programming to recognize plans
[Rich and Waters-90]
Explanation-based constraint programming
[Guéhéneuc and Antoniol-2008]
Use of queries
[Kullbach and Winter-99]
Generic fuzzy reasoning nets
[Niere et al.-2002].
Graph-transformation techniques
[Tsantalis et al.-2006]
CSMR 2011
Oldenburg](https://image.slidesharecdn.com/csmr11b-ppt-130410095510-phpapp02/85/CSMR11b-ppt-3-320.jpg)
![Related work (II)
Similar to our approach: [Tonella and Antoniol-01] in which concept
analysis was used to infer domain-specific design patterns.
Main Difference :
“maximal collections of class sets having the same relations
between them” VS isomorphic induced subgraphs of fixed size
improves on scalability via an efficient sub-graph enumeration
eliminates the need of manual inspection of “concept lattice”.
Inspiration from sub-graph mining algorithms
Sub-graph discovery problem : discover all recurring sub-graphs, or
only the most frequent ones.
Limitations of previous works
(restriction to undirected graphs, or to sub-graphs with limited
number of labels on their arcs [Wernicke and Rasche-2006] )
Our own tool SGFinder
CSMR 2011
Oldenburg](https://image.slidesharecdn.com/csmr11b-ppt-130410095510-phpapp02/85/CSMR11b-ppt-4-320.jpg)
![Case Study - Objects
Rhino: JavaScript/ECMAScript interpreter and compiler
• Defect data extracted from [Eaddy et al-2008] study
• Change data mined from the Rhino CVS logs.
ArgoUML: UML CASE tool written in Java.
• Defect data from the ArgoUML customized Bugzilla repository
• Change data from the SVN logs.
Class diagrams recovered using the Ptidej tool suite and its PADL
meta-model.
Size of micro-architectures of interest: 3, 4 and 5 nodes.
CSMR 2011
Oldenburg](https://image.slidesharecdn.com/csmr11b-ppt-130410095510-phpapp02/85/CSMR11b-ppt-11-320.jpg)
![References
[Guéhéneuc and Antoniol-2008] Y.-G. Guéhéneuc and G. Antoniol,
“DeMIMA: A multi-layered framework for design pattern identification,”
Transactions on Software Engineering (TSE), vol. 34, no. 5, pp. 667–684,
September 2008.
[Tonella and Antoniol-01] P. Tonella and G. Antoniol, “Inference of object
oriented design patterns,” Journal of Software Maintenance - Research and
Practice, vol. 13, no. 5, pp.309–330, September-October 2001.
[Rich and Waters-90] C. Rich and R. C. Waters, The Programmer’s Apprentice,
1st ed. ACM Press Frontier Series and Addison-Wesley, January 1990.
[Kullbach and Winter-99] B. Kullbach and A. Winter, “Querying as an
enabling technology in software reengineering,” in Proceedings of the 3rd
Conference on Software Maintenance and Reengineering, P. Nesi and C. Verhoef,
Eds. IEEE Computer Society Press, March 1999, pp. 42–50.
[Niere et al.-2002] J. Niere, W. Sch¨afer, J. P. Wadsack, L. Wendehals, and J.
Welsh, “Towards pattern-based design recovery,” in Proceedings of the 24th
International Conference on Software Engineering, M.Young and J. Magee, Eds.
ACM Press, May 2002, pp. 338–348.
CSMR 2011
Oldenburg](https://image.slidesharecdn.com/csmr11b-ppt-130410095510-phpapp02/85/CSMR11b-ppt-24-320.jpg)
![References
[Tsantalis et al.-2006] N. Tsantalis, A. Chatzigeorgiou, G. Stephanides, and S.
Halkidis, “Design pattern detection using similarity scoring,” Transactions on
Software Engineering, vol. 32, no. 11, November 2006.
[Wernicke and Rasche-2006] S. Wernicke and F. Rasche, “A tool for fast
network motif detection,” Bioinformatics, vol. 22, pp. 1152–1153, 2006.
[Eaddy et al.-2008] M. Eaddy, T. Zimmermann, K. D. Sherwood,V. Garg, G. C.
Murphy, N. Nagappan, and A.V. Aho, “Do crosscutting concerns cause
defects?” IEEE Transactions on Software Engineering, vol. 34, no. 4, pp. 497–515,
2008.](https://image.slidesharecdn.com/csmr11b-ppt-130410095510-phpapp02/85/CSMR11b-ppt-25-320.jpg)
CSMR11b.ppt
- 1. Sub-graph Mining: Identifying Micro-architectures in Evolving Object-oriented Software Ahmed Belderrar, Segla Kpodjedo, Yann-Gaël Guéhéneuc, Giuliano Antoniol, Philippe Galinier CSMR 2011 Oldenburg
- 2. Context & Motivation Coding activity Active Research Design Good? OO Patterns application Known Anti Bad? Patterns New organizations of classes and Almost Nothing relations Micro-Architectures UnKnown Ugly? Preliminary study exploring the feasibility and prospects of an approach CSMR 2011 aimed at building empirical knowledge about organizations of classes. Oldenburg
- 3. Related work (I) In general, the proposed approaches rely on a library of known motifs and use architectural recovery techniques based on sub-graph matching or on motifs properties. Constraint Programming to recognize plans [Rich and Waters-90] Explanation-based constraint programming [Guéhéneuc and Antoniol-2008] Use of queries [Kullbach and Winter-99] Generic fuzzy reasoning nets [Niere et al.-2002]. Graph-transformation techniques [Tsantalis et al.-2006] CSMR 2011 Oldenburg
- 4. Related work (II) Similar to our approach: [Tonella and Antoniol-01] in which concept analysis was used to infer domain-specific design patterns. Main Difference : “maximal collections of class sets having the same relations between them” VS isomorphic induced subgraphs of fixed size improves on scalability via an efficient sub-graph enumeration eliminates the need of manual inspection of “concept lattice”. Inspiration from sub-graph mining algorithms Sub-graph discovery problem : discover all recurring sub-graphs, or only the most frequent ones. Limitations of previous works (restriction to undirected graphs, or to sub-graphs with limited number of labels on their arcs [Wernicke and Rasche-2006] ) Our own tool SGFinder CSMR 2011 Oldenburg
- 5. Our approach – Introduction Considering a class diagram as a labeled graph, we define a micro- architecture as the connected subgraph induced by a given subset of classes. Pre-requisite We model a class diagram as a labeled graph, with nodes being classes (and interfaces) and arcs representing the relations among classes. Arc label type of relation between two classes (e.g., association, aggregation, or inheritance). Example 1: Association 2: Aggregation/Composition 3: Inheritance CSMR 2011 Figure 1 A sub-graph G(V,E) extracted from Rhino Oldenburg (release 1.7R1).
- 6. Our approach – Definitions Labeled graph: given a set of labels L, it is defined by a triplet G = (V,A, l) where V is the vertex set of G, A ⊆ V × V the arc set, and l : A L the labeling function. Connected Graph: There is a chain between any two vertices. Isomorphic Graphs: There exists a perfect one-to-one correspondence between the vertices and edges of those graphs. Induced Subgraph: GX induced by X (X⊆V) is a graph which vertex set is X and which arc set contains all the arcs of G that link two vertices of X. Embedding of a sub-graph H of G: set X of vertices such that GX is isomorphic to H. Example Figure 2 A sub-graph G=(V,E) extracted from Rhino (release 1.7R1) with two embeddings CSMR 2011 of size three {7, 1, 3} and {2, 4, 5}. Oldenburg
- 7. Our approach – Algorithm (I) Inputs A graph G and the size k of the sought sub-graphs Output The set of induced subgraphs with their embeddings. Principle Generate all k-subsets X of V such that GX is connected. Generates only a limited number of k-subsets of V Underlying idea if GX is a connected induced sub-graph of G, then X can not contain two vertices x and y such that dist(x, y) ≥ k. CSMR 2011 Oldenburg
- 8. Our approach – Algorithm (II) SGFinder Build k disjoint subsets of vertices, denoted by L0...Lk−1 (line 3-5), where each Li contains the vertices which distance to x equals i (L0 = {x}) To build X (line 6, developed in the CompleteSG() procedure), we choose the unique vertex present in L0 (namely x), plus one or more vertices chosen in L1, plus one or more vertices chosen in L2, and so on until we reach a total number k of vertices. CSMR 2011 Oldenburg
- 9. Our approach – Algorithm (III) Illustration Figure 1 A sub-graph G(V,E) extracted from Rhino (release 1.7R1). - Let us choose k = 4, - And let us assume that the first chosen vertex is x = 0. 1. We have L0 = {0}, L1 = {3}, L2 = {1, 2}, and L3 = {4, 6, 7}. 2. There are now seven possibilities for X: {0, 3, 1, 2}, {0, 3, 1, 4}, {0, 3, 1, 6}, {0, 3, 1, 7}, {0, 3, 2, 4}, {0, 3, 2, 6}, {0, 3, 2, 7}. 3. Some of the induced sub-graphs are not connected, e.g., G{0,3,1,4} and are discarded. CSMR 2011 Oldenburg
- 10. Case Study Goal Identify in OO systems, micro-architectures with desirable or harmful properties. Quality focus Verify applicability of SGFinder to small and medium size OO programs in order to detect micro-architectures that are stable or fault-prone. Perspective Researchers, developers, and managers, who want to get information about undocumented micro-architectures found in OO systems and possible related drawbacks. Context Two open-source systems: the Rhino JavaScript/ECMAScript interpreter, and the ArgoUML CASE tool. CSMR 2011 Oldenburg
- 11. Case Study - Objects Rhino: JavaScript/ECMAScript interpreter and compiler • Defect data extracted from [Eaddy et al-2008] study • Change data mined from the Rhino CVS logs. ArgoUML: UML CASE tool written in Java. • Defect data from the ArgoUML customized Bugzilla repository • Change data from the SVN logs. Class diagrams recovered using the Ptidej tool suite and its PADL meta-model. Size of micro-architectures of interest: 3, 4 and 5 nodes. CSMR 2011 Oldenburg
- 12. Case Study – Research Questions RQ1 – Does SGFinder scale up to medium size programs and what kind of microarchitectures are found in OO systems? RQ2 – Are there micro-architectures particularly fault- prone or fault-free? RQ3 – Are there micro-architectures particularly stable or change-prone? CSMR 2011 Oldenburg
- 13. Case Study – Analysis Method (I) Characterizing a micro-architecture mAi Connectivity •Total number of relations (nbRel), •Number of associations (nbAssoc), •Number of aggregations or compositions (nbAggr), •Number of inheritances (nbInher), •Number of ”cyclic relations” (nbCycl). Presence and repartition •Number of releases in which mAi can be found (nbReleases) •Number of embeddings of mAi in a given release (nbEmbeddings) •Number of zones (nbZones). an embedding is counted only when it does not share any common edge with a previously found embedding •Number of classes appearing at least once in mAi (nbClasses) CSMR 2011 Oldenburg
- 14. Case Study – Analysis Method (II) Answering the Research Questions RQ1 Execution times and Characterization of micro-architectures RQ2 and RQ3 1. For each mA, we compute the percentage of its classes that are fault-prone (i.e., with one or more documented faults) or changed between two subsequent releases. 2. Rank micro-architectures from the fault(change)-prone to the fault(change)-free using the precision measure of step 1. 3. Inspection of the top 10% and bottom 10% of ranked microarchitectures hints of what makes those micro- architectures outstanding. Use of descriptive statistics : the five-number summary statistic (Min,Quartile1,Median,Quartile3,Max). CSMR 2011 Oldenburg
- 15. Results – RQ1.1 SGFinder Applicability Computation times: From a few seconds to ...two days and half. 3-nodes micro-architectures 20 seconds at most 4-nodes micro-architectures less than 30 minutes 5-nodes ...another story Rhino max <8 min Argo max > 60 h to retrieve the 82,877 different 5-nodes mAs in ArgoUML 0.17.5 and their 13,741,073,588 embeddings. Depend mostly on the edge density of the considered class diagrams. More specifically, the single most time-costly factor is the presence of highly connected nodes which lead to a near-combinatorial explosion of the number of embeddings. CSMR 2011 Oldenburg
- 16. Results – RQ1.2 Micro-architectures’ Description Number of different micro-architectures. - 3 basic relations (association 1, aggregation 2 and inheritance 3) - 4 derived mixed cases (12,13,23,123) - No relation 8 possible connections between two nodes For k × k pairs of nodes (including loops), if we do not take into account symmetry At most 8 (k×k−(k−1)) ×7k−1 connected subgraphs of k nodes. For k=5, in theory, possibly 22 × 1021 different micro-architectures. In practice, the union of five-nodes micro-architectures from Rhino and Argo contains only about 32 × 104 different micro-architectures. Suggests, unsurprisingly, that the set of micro-architectures is CSMR 2011 restricted to specific subgraphs Oldenburg
- 17. Results – RQ1.2 Micro-architectures’ Description Characterization Averages over the two systems CSMR 2011 Oldenburg
- 18. Results – RQ1.2 Micro-architectures’ Description CSMR 2011 Illustration - Most Connected 5-nodes micro-architectures Oldenburg
- 19. Results – RQ2 Fault proneness Figure 4. Example of a particularly faulty (precision=81) 4-nodes micro-architecture (almost) every class ”‘talks”’ to every other class. In fact, when we retrieve the set of micro-architectures with three cyclic relations and where every class is connected to every other class, we obtain a 5- number summary (28,44,56,67,81) with values mostly higher to those from the top 10% most faulty CSMR 2011 microarchitectures (42,45,48,54,83). Oldenburg
- 20. Results – RQ3 Change proneness Figure 5. Example of a particularly stable (precision=30) 4-nodes micro-architecture presents a ”cascade”- like pattern and we were able to retrieve many other similar organizations with mostly low changeability. CSMR 2011 Oldenburg
- 21. Threats to Validity Threats to conclusion validity concern the relationship between the treatment and the outcome. We do not claim any causal relation between the microarchitectures and unwanted or desired features. Threats to external validity concern the possibility of generalizing our results. The study is limited to 2 systems: Rhino and ArgoUML. But threats are mitigated given that the two systems correspond to different domains and applications, have different sizes, and are developed by different teams. CSMR 2011 Oldenburg
- 22. Conclusion I presented SGFinder, an algorithm and a tool to support micro- architecture discovery based on a reformulation as a sub-graph mining problem. SGFinder uses an effective enumeration technique that allows us to infer instances of micro-architectures and to study micro- architecture properties such as stability or fault proneness. We used SGFinder on 8 releases of 2 well known Java applications: Rhino and ArgoUML. We provided insight about the kind of micro-architectures (of 3, 4 or 5 nodes) found in OO systems. We characterize the most and least faulty/changed micro- architectures and report some of the most interesting micro- architectures with respect to their connectivity and frequency. CSMR 2011 Oldenburg
- 23. Future Ongoing Work Despite the encouraging results, more work is needed to (i) further optimize the proposed algorithm and gain in scalability, and/or investigate the partitioning of large OO diagrams into subsystems (ii) define heuristics and rules able to classify micro-architectures discovered in a system under development, (iii) go beyond the micro-architectures and analyze, similarly to design patterns, the roles played by participant classes, and (iv) provide qualitative analysis and validation of the findings. For generalization purposes, our plan for future work also include replication of the study on a larger system. CSMR 2011 Oldenburg
- 24. References [Guéhéneuc and Antoniol-2008] Y.-G. Guéhéneuc and G. Antoniol, “DeMIMA: A multi-layered framework for design pattern identification,” Transactions on Software Engineering (TSE), vol. 34, no. 5, pp. 667–684, September 2008. [Tonella and Antoniol-01] P. Tonella and G. Antoniol, “Inference of object oriented design patterns,” Journal of Software Maintenance - Research and Practice, vol. 13, no. 5, pp.309–330, September-October 2001. [Rich and Waters-90] C. Rich and R. C. Waters, The Programmer’s Apprentice, 1st ed. ACM Press Frontier Series and Addison-Wesley, January 1990. [Kullbach and Winter-99] B. Kullbach and A. Winter, “Querying as an enabling technology in software reengineering,” in Proceedings of the 3rd Conference on Software Maintenance and Reengineering, P. Nesi and C. Verhoef, Eds. IEEE Computer Society Press, March 1999, pp. 42–50. [Niere et al.-2002] J. Niere, W. Sch¨afer, J. P. Wadsack, L. Wendehals, and J. Welsh, “Towards pattern-based design recovery,” in Proceedings of the 24th International Conference on Software Engineering, M.Young and J. Magee, Eds. ACM Press, May 2002, pp. 338–348. CSMR 2011 Oldenburg
- 25. References [Tsantalis et al.-2006] N. Tsantalis, A. Chatzigeorgiou, G. Stephanides, and S. Halkidis, “Design pattern detection using similarity scoring,” Transactions on Software Engineering, vol. 32, no. 11, November 2006. [Wernicke and Rasche-2006] S. Wernicke and F. Rasche, “A tool for fast network motif detection,” Bioinformatics, vol. 22, pp. 1152–1153, 2006. [Eaddy et al.-2008] M. Eaddy, T. Zimmermann, K. D. Sherwood,V. Garg, G. C. Murphy, N. Nagappan, and A.V. Aho, “Do crosscutting concerns cause defects?” IEEE Transactions on Software Engineering, vol. 34, no. 4, pp. 497–515, 2008.
- 26. Thanks for the attention Questions? CSMR 2011 Oldenburg