the technical impact of social links in free software development rishab aiyer ghosh [email protected]...
TRANSCRIPT
![Page 1: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/1.jpg)
The technical impact of social links The technical impact of social links in free software developmentin free software development
Rishab Aiyer [email protected]
MERIT/Infonomics, University of Maastricht
Oxford Workshop on Libre Software, Oxford Internet Institute, June 25, 2004
![Page 2: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/2.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Some data on the Linux kernelSome data on the Linux kernel
-number of developers,
share of sub-modules
![Page 3: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/3.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Some data on the Linux kernelSome data on the Linux kernel
-number of sub-modules per author
![Page 4: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/4.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Some data on the Linux kernelSome data on the Linux kernel
-number of co-authors per author
![Page 5: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/5.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Social choice of modules?Social choice of modules?
Most modules have just a few authors
Most authors have contributed to just one module
However, these lone contributions are not made to 1-author modules
New contributors choose modules with many other contributors
![Page 6: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/6.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Mean developers per moduleMean developers per module
100.00 200.00 300.00 400.00
avcodv25 IF nz25=1 (else MISSING)
0%
5%
10%
15%
Per
cen
t
Mean number of “co-developers” per module for 1-module contributors: over 45% have more than 200 co-developers per module.
![Page 7: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/7.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Mean developers per moduleMean developers per module
Mean number of “co-developers” per module:
- for 1-module contributors: over 45% have more than 200 co-developers per module.
- for 2-module contributors (red overlay), only a slight change. I.e. 2nd modules are not much smaller than 1st modules contributed to.
![Page 8: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/8.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Mean developers per moduleMean developers per module
Correlation: number of modules authored by average number of developers per module
avdevv1.0 (n=158) -0.225v2.0.30 (n=618) -0.222v2.5.25 (n=2263) -0.145
(Pearson 2-tailed: correlations are significant at the 0.01 level)
![Page 9: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/9.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Developer numbers / module sizeDeveloper numbers / module size
Contributors don’t necessarily know how many co-authors they have / will have for a given module
Code size of module is easily available Size is an explicit proxy for “importance” Hypothesis: developers are attracted to
“important” projects, partly because they have many other developers
![Page 10: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/10.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Developer numbers / module sizeDeveloper numbers / module size
Module size highly correlated to number of authors
Correlation
v1.0 (n=30) 0.890
v2.0.30 (n=60) 0.892
v2.5.25 (n=169) 0.894(Pearson 2-tailed: correlations are significant at the 0.01 level)
![Page 11: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/11.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Module associationModule association
Modules can be linked together to form a graph based on at least two criteria
Authorship: one or more contributors are common to modules
Code: one module depends on functions defined in another module
It turns out that the co-incidence of these two attributes is quite high
![Page 12: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/12.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Module association: Module association: authorsauthors, v1.0, v1.0
![Page 13: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/13.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Module association: Module association: codecode, v1.0, v1.0
![Page 14: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/14.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Module association: Module association: bothboth, v1.0, v1.0
![Page 15: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/15.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Module association: Module association: bothboth, , v2.5.25v2.5.25
![Page 16: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/16.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Module association: Module association: bothboth, , v2.5.25v2.5.25
![Page 17: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/17.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Exploring the author-code linkExploring the author-code link
Strong degree of coincidence – between author link and code link
Significant correlation between strength of author link and presence of code link
Indications of dynamic impact of author link on future code links (and vice versa)
![Page 18: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/18.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Exploring the author-code linkExploring the author-code link
Phi 4-point similarity between binary variables for author link & code link:
v1.0 (n=435) 0.122
v2.0.30 (n=1770) 0.254
v2.5.25 (n=14196) 0.341
![Page 19: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/19.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Exploring the author-code linkExploring the author-code link
Spearman’s rho: scalar variables for strength of author link against binary variable for presence of code link:
v1.0 (n=435) 0.130
v2.0.30 (n=1770) 0.241
v2.5.25 (n=14196) 0.341(strength measured by number of common authors;
also tested with other strength measures)
![Page 20: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/20.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Exploring the author-code linkExploring the author-code link
Analytical problems: no good fit model with regression, possibly due to highly skewed data
Hard to select strength (rather than binary presence) variable for code dependency link
Time lag between versions too big for dynamic analysis
![Page 21: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/21.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Future stepsFuture steps
Similar exercise using finer dynamic granularity (and possibly CVS data, for activity rather than cumulative authorship) may allow better interpretation
Better data on code dependency (especially dependency strength) may help in identifying relationships
![Page 22: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/22.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Tentative conclusionsTentative conclusions
Significant relationship between social and technical links between modules
Direction of causality is unclear Impact on code development is, however,
potentially very high “tip of the iceberg” – e.g. code reuse
(including dependency on “trivial” rather than “complex” functions) may be much higher with greater social links
![Page 23: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/23.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Future stepsFuture steps
Use of clustering algorithms may help identify predictor relationships
Common authorship may predict code dependency group-wise more than pair-wise
![Page 24: The technical impact of social links in free software development Rishab Aiyer Ghosh rishab@dxm.org MERIT/Infonomics, University of Maastricht Oxford Workshop](https://reader035.vdocuments.net/reader035/viewer/2022062422/56649ef65503460f94c092da/html5/thumbnails/24.jpg)
Oxford, June 25, 2004 © Rishab Aiyer Ghosh / MERIT
Further informationFurther information
www.flossproject.orgCODD technical papers:
orbiten.org/codd/
codd.berlios.de