A Formal Specification for a Real-Time Train Controller

We give a formal specification for a real-time controller for trains that operate on the Italian railway network. The controller will control train movements and is part of a larger system destined to guarantee safety with respect to dangers originating from train traffic in the railway network. Based on an informal specification document from the Italian railway company, we construct a simple state-based model and formalise it in terms of the property-based specification language TRIO. The obtained specification being formal, we are able to perform certain verifications on it, such as checking its satisfiability and verifying correctness of refinement steps.

Using the COOPN/2 Formal Method for Groupware Systems Engineering

COOPN/2 (Concurrent Object Oriented Petri Nets) is an object-oriented specification formalism based on Petri Nets and algebraic specifications. It is fully adapted to the specification of complex concurrent systems because it inherits the advantages of algebraic specifications for the specification of abstract data types, from Petri nets to describe partially ordered events and from the object-oriented paradigm for the 'structuring' of the specification. All these features are especially useful when speaking about groupware systems, which are application which demand rich modeling capabilities on concurrency and structuring. This paper explains the major principles for the specification of groupware systems with COOPN/2. This is done by mapping groupware concepts into COOPN/2 concepts and is demonstrated by means of an example of shared distributed diaries. Furthermore, we present the work done on the methodological approach for the development of concurrent object-oriented systems with COOPN/2 which cover analysis, specification, design, prototyping and testing. groupware, replication, object-orientation, distributed systems, formal methods, Petri Nets, algebraic specification of abstract data types, software engineering. Keywords : groupware, replication, object-orientation, distributed systems, formal methods, Petri Nets, algebraic specification of abstract data types, software engineering.

1996

Verification of Software Network Functions with No Verification Expertise

Arseniy Zaostrovnykh

Software network functions (NFs), such as a network address translator, load balancer, or proxy, promise to bring flexibility and rapid innovation to computer networks and to reduce operational costs. However, continuous updates and flexibility typically come with the risk of bugs. As networks are a critical component of modern computing (e.g., the Internet), both users and operators demand that they be as reliable as economically possible. One way to guarantee reliability is by formal verification. To prove the absence of bugs, formal verification applies rigorous mathematical logic. State-of-the-art formal verification methods either require substantial effort and verification expertise on the part of the practitioner or lack completeness. In particular, theorem proving, to guide a mechanical proof checker, requires a human to annotate the code they verify with logical steps, and this annotation is something for which NF developers typically lack the time or expertise to do. Whereas, symbolic execution (SE) can analyze the code automatically, hence requiring significantly less effort; but it has difficulties with the so-called path explosion. Fortunately, software NFs have a special structure:

They feature a single infinite event-processing loop, and most of their code is SE-friendly.
The non-SE-friendly parts implement well-defined data structures that are reused across NFs.

In this dissertation, we present Vigor, a new method and tool for verifying software NFs; it exploits this special structure of NFs. We make two contributions: First, we show that it is possible to develop software NFs that are guaranteed to satisfy a formal specification and at the same time exhibit competitive performance. We develop a practical toolchain for developing NFs in C by using a standard fast packet-processing framework. Vigor provides a library of formally verified data structures (libVig); this library replaces the parts posing a challenge to SE. The Vigor toolchain applies exhaustive SE to the NF code that uses libVig. Vigor introduces a new technique, which we call "lazy proofs", for automatically stitching the proofs together and, ultimately, for certifying the whole NF correct relative to its functional specification. Second, we show that practical high-level semantic verification of software NFs is possible. By practical, we mean a verification that (a) requires no verification input other than the specification (push-button), (b) guarantees correctness of the entire runtime software stack, i.e., OS, driver, framework, and NF (full-stack), and (c) does not require writing a comprehensive specification to verify the first line of code (pay-as-you-go). In summary, Vigor provides push-button, full-stack, pay-as-you-go formal verification of software NFs with no reduction in NF performance or developer productivity. To substantiate our claims we implement five NFs, prove their correctness, and show that their performance is as good as the performance of the fastest non-verified software NFs. All of our NFs achieve a median latency of 4.1±0.2 us and over 4 Mpps throughput with 10 Gbps network interfaces. In terms of developer productivity, an average partial property takes 6.4 lines of Python specification, and the verification of the full stack (NF, packet-processing framework, OS, driver) takes up to one day. The Vigor toolchain is open-source and is available, along with tutorials and example NF implementations, at vigor.epfl.ch.

EPFL2020

Verification of Software Network Functions with No Verification Expertise

Arseniy Zaostrovnykh

They feature a single infinite event-processing loop, and most of their code is SE-friendly.
The non-SE-friendly parts implement well-defined data structures that are reused across NFs.

EPFL2020