Synthesis of Fault-Tolerant Reliable Broadcast Algorithms With Reinforcement Learning

Fault-tolerant algorithms, such as Reliable Broadcast, assure the correct operation of modern distributed systems, even when some of the system nodes fail. However, the development of distributed algorithms is a manual and complex process, where slight changes in requirements can require a complete redesign of the algorithm. To automate the process of developing such algorithms, this work presents a new approach that uses Reinforcement Learning to synthesize correct and efficient fault-tolerant distributed algorithms. This work shows the first application of the approach on the synthesis of fault-tolerant Reliable Broadcast algorithms. The presented technique is capable of synthesizing correct and efficient algorithms with the same performance as others available in the literature as well as a new Byzantine tolerant algorithm, in only 12,000 learning episodes. Based on the success of this implementation, we aim, in the future, to extend this technique to other distributed algorithms such as Consensus.