Modularity for decidability of deductive verification with applications to distributed systems

Abstract

Proof automation can substantially increase productivity in formal verification of complex systems. However, unpredictablility of automated provers in handling quantified formulas presents a major hurdle to usability of these tools. We propose to solve this problem not by improving the provers, but by using a modular proof methodology that allows us to produce decidable verification conditions. Decidability greatly improves predictability of proof automation, resulting in a more practical verification approach. We apply this methodology to develop verified implementations of distributed protocols, demonstrating its effectiveness. CCS Concepts • Software and its engineering → Formal software verification;

Pretend synchrony: synchronous verification of asynchronous distributed programs

“…We have used Goolong to develop four challenging case studies: the classic two-phase commit protocol, the Raft leader election protocol, single-decree Paxos protocol and a Multi-Paxos based distributed key-value store that employs our protocol rounds. In our tests, the key-value store outperformed other verified stores [Drăgoi et al 2016;Taube et al 2018a] while staying within 3x of an unverified state-of-the-art implementation [Moraru et al 2013]. Goolong Library To be able to extract IceT programs from programs written in Go, we wrote a library that exposes IceT language primitives (e.g., sequential iteration, symmetric processes, send and recvTO etc) in Go.…”

“…For this, we run it on three separate Amazon EC2 t2.micro instances in the same availability zone using a client executing a series of PUT or GET queries. Table Table 2 show the throughput compared to Psync [Drăgoi et al 2016], Ivy [Taube et al 2018b], and IronFleet [Hawblitzel et al 2015a]. While we were able to run Psync, we were unable to run Ivy and IronKV and instead we provide their published results as a (albeit not directly comparable) reference.…”

“…Disel [Sergey et al 2018;Wilcox et al 2017] aims to enable the user to compose protocols in an effort to modularize proofs of correctness. Similarly, [Taube et al 2018a] aims to modularize verification of distributed algorithms by splitting verification conditions into subsets each of which uses a decidable fragment of first-order logic, were different subsets may use different fragments. [Padon et al 2017] presents a method of (manually) transforming general, first-order verification conditions into effectively propositional logic, and applies it to verifying several Paxos variants.…”

Verifying message-passing programs with dependent behavioural types

“…Outside the realm of process calculi, various works tackle the problem of protocol-aware verification, e.g., [40,71,74]. We share similar goals, although we adopt a different theory and design, leading to different tradeoffs: crucially, the works above develop new languages, or build upon a powerful dependently-typed host language (Coq) with interactive proofs, to support rich representations of protocol state.…”

Asphalion: trustworthy shielding against Byzantine faults

“…Ivy [Padon et al 2016] initially supported debugging infinite-state systems using bounded verification, and verifying their safety by gradually building inductive invariants. Their decidable decomposition notion [Taube et al 2018] (i.e., systems, models and proofs must be built modularly to enable the use of different decidable logics) allowed Ivy to automatically verify the correctness of implementations of crash-fault tolerant systems such as Raft and Paxos (as opposed to models in [Padon et al 2017]). Ivy also supports liveness by reducing it to safety .…”