PhD thesis: Abstractions for causal analysis and explanations in concurrent programs

Within the ANR project “DCore – Causal Debugging for Concurrent Systems” we are looking for a highly motivated PhD student. The first goal of the thesis is to investigate the use of abstractions for the construction of causal explanations. During the last couple of years, several approaches have been proposed to construct, from the execution of a system violating an expected property P, concise explanations why P was violated, and which components or events caused this violation, see e.g. [JRS04,BBD+12,DF+12,FMN15,GLM15,GS15]. However, these approaches either were limited to small systems, or used ad-hoc approximations to make them scale. The goal of the DCore project is to develop and implement causal analysis for a full-fledged programming language, in order to answer questions of the form “what would have been the outcome if component A had satisfied its specification?” or “which component faults were necessary to entail the observed effect?”.

We are therefore interested in developing abstractions that “compose well” with causal analyses, and understanding precisely how explanations found on the abstraction relate to explanations on the concrete system. It is worth noting that the presence of abstraction, which inherently comes with some induction and extrapolation processes, completely recasts the issue of reasoning about causality. Causal traces do no longer describe only potential
scenarios in the concrete semantics, but also mix some approximation steps coming from the computation of the abstraction itself. Causal explanation is then intertwined with the issue of alarm diagnosis [R05], and not all explanations are replayable counter-examples: they may contain some steps witnessing some lack of accuracy in the analysis. Vice versa, a research question to be addressed is how to define causal analyses that have a well understood behavior under abstraction. The second goal of the thesis is to implement and validate the theoretical developments as part of a causal debugger for a full-fledged programming language, such as Java with actors, or Erlang.

Candidates should have a good background in formal methods. Good programming skills are required.

The starting date is flexible.

The PhD thesis will be co-advised by Jérôme Feret (Antique team, INRIA Paris) and Gregor Gössler (Spades team, INRIA Grenoble). Please do not hesitate to contact us for more information: firstname.lastname@inria.fr.

References:

[BBD+12] I. Beer, S. Ben-David, H. Chockler, A. Orni, and R.J. Trefler. Explaining counterexamples using causality. FMSD 40(1), pp. 12-40, 2012.

[DF+12] V. Danos, J. Feret, W. Fontana, R. Harmer, J. Hayman, J. Krivine, C. Thompson-Walsh, and G. Winskel: Graphs, Rewriting and Pathway Reconstruction for Rule-Based Models. Proc. FSTTCS 2012, LIPICS 18, 2012.

[FMN15] T. Ferrère, O. Maler, D. Nickovic: Trace Diagnostics Using Temporal Implicants. Proc. ATVA 2015: 241-258, 2015.

[GLM15] G. Gössler and D. Le Métayer: A general framework for blaming in component-based systems. Sci. Comput. Program. 113(3):223-235, 2015.

[GS15] G. Gössler and J.-B. Stefani: Fault Ascription in Concurrent Systems. Proc. TGC’15, LNCS 9533, 2015.

[JRS04] H. Jin, K. Ravi, and F. Somenzi: Fate and free will in error traces. STTT 6(2):102-116, 2004.

[R05] X. Rival. Abstract Dependences for Alarm Diagnosis. Proc. APLAS 2005, LNCS 3780, 347-363, 2005.

Comments are closed.