Automated Synthesis of Distributed Controllers [article]

Anca Muscholl
2015 arXiv   pre-print
Synthesis is a particularly challenging problem for concurrent programs. At the same time it is a very promising approach, since concurrent programs are difficult to get right, or to analyze with traditional verification techniques. This paper gives an introduction to distributed synthesis in the setting of Mazurkiewicz traces, and its applications to decentralized runtime monitoring. 1 Context Modern computing systems are increasingly distributed and heterogeneous. Software needs to be able to
more » ... exploit these advances, providing means for applications to be more performant. Traditional concurrent programming paradigms, as in Java, are based on threads, shared-memory, and locking mechanisms that guard access to common data. More recent paradigms like the reactive programming model of Erlang [4] and Scala [35,36] replace shared memory by asynchronous message passing, where sending a message is non-blocking. In all these concurrent frameworks, writing reliable software is a serious challenge. Programmers tend to think about code mostly in a sequential way, and it is hard to grasp all possible schedulings of events in a concurrent execution. For similar reasons, verification and analysis of concurrent programs is a difficult task. Testing, which is still the main method for error detection in software, has low coverage for concurrent programs. The reason is that bugs in such programs are difficult to reproduce: they may happen under very specific thread schedules and the likelihood of taking such corner-case schedules is very low. Automated verification, such as model-checking and other traditional exploration techniques, can handle very limited instances of concurrent programs, mostly because of the very large number of possible states and of possible interleavings of executions. Formal analysis of programs requires as a prerequisite a clean mathematical model for programs. Verification of sequential programs starts usually with an abstraction step -- reducing the value domains of variables to finite domains, viewing conditional branching as non-determinism, etc. Another major simplification consists in disallowing recursion. This leads to a very robust computational model, namely finite-state automata and regular languages. Regular languages of words (and trees) are particularly well understood notions. The deep connections between logic and automata revealed by the foundational work of B\"uchi, Rabin, and others, are the main ingredients in automata-based verification .
arXiv:1506.02369v1 fatcat:effftlf33jb2ninmahnp53uk7q