Functional Analysis of Large Software Networks Natasa Przulj, Gordon Lee and Igor Jurisica IBM CAS, University of Toronto.

Functional Analysis of Large Functional Analysis of Large Software NetworksSoftware Networks

Natasa Przulj, Gordon Lee and Igor JurisicaIBM CAS, University of Toronto

Overview:Overview:

Software Call GraphsModels of Large NetworksProperties of a Software Network (PSQL):

Preliminary ResultsBugs versus Network PropertiesFuture Work

Software Call GraphsSoftware Call Graphs

PSQL 7.3 Call Graph (R. Holt, J. Wu)

Models of Large NetworksModels of Large Networks Random Graphs (Erdos, Renyi) Generalized Random Graphs (Bander, Canfield) Small-World (Watts, Strogatz; Newman, Watts) Scale-Free (Simon; Barabasi, Albert, Jeong)

Strogatz, Nature,

410

Properties of Large Networks:Properties of Large Networks: Degree distribution Diameter Clustering

Degree distrib. ex. (NSW, Phys Rev E, 64):

# Nodes of PSQL 7.0, 7.1, 7.2, 7.3: 4639, 5127, 5568, 5996

# Edges of PSQL 7.0, 7.1, 7.2, 7.3: 17010, 19790, 21616, 23802

Degree Distribution Shortest Path Length Distribution

Graph PropertiesGraph Properties

Bug ReportsBug Reports•PSQL: no proper database of bug reports (severity…) - limitation•Used Google Web APIs to search (Ben Vitale)•http://archives.postgresql.org/, http://developer.postgresql.org/

With Bug Reports

Lines of Code (LOC)Lines of Code (LOC)

LOC Statistics for graph groups for PSQL 7.3:

Future WorkFuture Work

Properties of SW call graphs to aid identifying buggy modules

SW Design versus Network StructureOverlap of network clusters with “natural

modules”CAS interested in similar analysis on DB2

Thanks:Thanks:

Igor Jurisica (supervisor, U of Toronto) Derek Corneil (supervisor, U of Toronto) Gordon Lee (IBM CAS) Ric Holt (U of Waterloo) Grad students: Jingwey Wu, Benjamin Vitale,

Wayne Hayes, Daniela Rosu, Cristiana Chitic, Travis Gagie, Robert Vracaric, Nina Przulj

IBM Center for Advanced Studies (CAS)