A Framework for Verifying Data-Centric Protocols

Abstract

Abstract. Data centric languages, such as recursive rule based languages, have been proposed to program distributed applications over networks. They simplify greatly the code, which is orders of magnitude shorter, much more declarative, while still admitting efficient distributed execution. We show that they also provide a promising approach to the verification of distributed protocols, thanks to their data centric orientation, which allows to explicitly handle global structures, such as the topology of the network, routing tables, trees, etc, as well as their properties. We consider a framework using an original formalization in the Coq proof assistant of a distributed computation model based on message passing with either synchronous or asynchronous behavior. The declarative rules of the Netlog language for specifying distributed protocols, as well as the virtual machines for evaluating these rules, are encoded in Coq as well. We consider as a case study tree protocols, and show how this framework enables us to formally verify them in both the asynchronous and synchronous setting.

Formal Verification of Netlog Protocols

“…As for [1], our results are supported by a Coq development available in [17]. The main differences and improvements with relation to [1] are as follows:…”

“…This greatly simplifies the code, while still admitting efficient distributed execution. From previous work [1], we know that such languages also provide a promising approach to the verification of distributed protocols.…”

“…We present here the encoding in Coq of the framework defined in [1], and we extend it in order to take non-monotonic features of Netlog into account. This includes a general library formalizing a distributed computation model based on message passing with either synchronous or asynchronous behavior, a representation of declarative rules of the Netlog language as well as their evaluation mechanism.…”

“…• We provide the intuitive ideas on the reasoning behind the (rather dry) proof of BFS given in [1], and make it more accurate using local reasoning. Let us emphasize that the use of a proof-assistant was of much help when analyzing our initial proof.…”

Messages with implicit destinations as mobile agents

“…They are more declarative, so facilitate programming, they parallelize well, so facilitate the execution, they manipulate explicitly data structures, so facilitate verification of their properties. Simple Netlog protocols for instance have already been verified [15] using the Coq proof assistant.…”