Section: New Results

Reductions between synchronous and asynchronous programming abstractions

 

Communication closed asynchronous protocols.

Participants : Andrei Damien, Cezara Drăgoi, Alexandru Militaru, Josef Widder.

 

Fault-tolerant distributed systems are implemented over asynchronous networks, where performance emerges from the load of the system. Due to asynchronous communication and the occurrence of faults (e.g., process crashes or the network dropping messages) the implementations are hard to understand and analyze. In contrast, synchronous computation models simplify design and reasoning.

In [17], we defined the first algorithm that automatically transforms an asynchronous protocol into a synchronous one. The method is sound but not complete. The transformation is based on an axiomatization of the notion of communication closure introduce by Elrad and Frances. If the asynchronous protocol is communication-closed then the translator will successfully compute its synchronous counter-part. Checking communication closure is done locally without considering any interferences between processes. The translator was successfully applied to Multi-Paxos, ViewStamped, and the atomic broadcast of Chandra and Toueg, generating the first synchronous counterparts of these protocols. The transformation from asynchronous to synchronous preserves the local states process go through and the exchanged messages. The translator has been implemented in a prototype tool called Athos, i.e., Asynchronous To Heard-Of Synchronizer, that is open source. The tool takes as input protocols in an intermediate protocol languages that has an asynchronous semantics and it is very close to C. These results have been published in one of the main verification venues Computer Aided Verification, CAV 2019 (acceptance rate <25% out of >250 submissions). The impact of the translator from asynchronous protocols to equivalent synchronous ones is important for the verification community because such a transformation reduces dramatically the state space and the set of traces to explore in order to prove the program correct, independently of the used verification technique.

Executable Rounds: a Programming Abstraction for Fault-Tolerant Protocols.

Participants : Cezara Drăgoi, Josef Widder, Damien Zufferey.

 

Fault-tolerant distributed systems are notoriously difficult to design and implement. Although programming languages for distributed systems is an active research area, appropriate synchronization primitives for fault-tolerance and group communication remains an important challenge. In [18] we present a new programming abstraction, HSync, for implementing benign and Byzantine distributed protocols. HSync is based on communication-closed rounds. Round models offer a simple abstraction for group communication and communication-closed rounds simplify dealing with faults. Protocols are implemented in a modular way in HSync. The language separates the message reception from the process local computation. It extends classic rounds with language constructs that give to the programmer the possibility to implement network and algorithm-specific policies for message reception. We have implemented an execution platform for HSync that runs on top of commodity hardware. We evaluate experimentally its performance, by comparing consensus implementations in HSync with LibPaxos3 and Bft-SMaRt, two consensus libraries tolerant to benign, resp. Byzantine faults.