Back to Search Start Over

A mathematical framework for design discovery from multi-threaded applications using neural sequence solvers

Authors :
Partha Pratim Das
Srijoni Majumdar
Amlan Chakrabarti
Nachiketa Chatterjee
Source :
Innovations in Systems and Software Engineering. 17:289-307
Publication Year :
2021
Publisher :
Springer Science and Business Media LLC, 2021.

Abstract

Comprehending existing multi-threaded applications effectively is a challenge without proper assistance. Research has been proposed to mine programs to extract aspects of high-level design but not much to reverse-engineer the concurrent design from multi-threaded applications. To address the same, we develop a generic mathematical model to interpret run-time non-deterministic events and encode functional as well as thread-specific behaviour in form of quantifiable features, which can be fitted into a standard solver for automated inference of design aspects from multi-threaded applications. We build a tool Dcube based on the mathematical model and use various classifiers of a machine learning framework to infer design aspects related to concurrency and resource management. We collect a dataset of 480 projects from Github, CodeProject and Stack Overflow and 3 benchmark suites—CDAC Pthreads, Open POSIX Test Suites and PARSEC 3.0 and achieve an accuracy score of around 93.71% for all the design choices.

Details

ISSN :
16145054 and 16145046
Volume :
17
Database :
OpenAIRE
Journal :
Innovations in Systems and Software Engineering
Accession number :
edsair.doi...........d84fa993d5cf49f83f262c4eea67726e
Full Text :
https://doi.org/10.1007/s11334-021-00393-8