The paper presents Rivulet, a distributed middleware for smart home applications. The main idea of Rivulet is to move from a state-of-the-art cloud-centric architectural solution, where the whole application logic is placed in the cloud, to a distributed home-centric solution where applications are distributed on the edge, represented by all the processing nodes integrated in the various smart appliances (such as TVs, refrigerators, ovens, and so on). The aim is to exploit the many “low-end” processing nodes distributed in the house to overcome the limitations of the cloud-centric solutions (connection latency and availability, architecture scalability, privacy, and so on). Moreover, another goal of the proposed middleware is to introduce a certain level of fault tolerance in the distributed architecture to overcome both transmission loss and node failures.

After a quick overall presentation of the middleware, the paper mainly focuses on three different aspects:

• The delivery service, presenting two different protocols: Gapless guarantees that all messages are eventually delivered (thus achieving fault tolerance in the communication), and Gap is a best-effort protocol. Such protocols can be selected based on the “criticality” of the considered application and necessity to not lose any message.
• The fault-tolerant execution scheme, based on the bully-based leader election algorithm, implements a sort of primary-secondary fault-tolerant approach capable of autonomously handling application crashes.
• A programming model, similar to the state-of-the-art solutions, offering the capability of writing event-driven dataflow applications.

The experimental campaign demonstrates the effectiveness of the delivery protocols and fault-tolerant execution scheme.

I appreciate how the authors address the problem and motivate their work. The paper presents a detailed overview of the literature on various topics related to the smart house scenario (smart house platforms, smart house case studies and applications, streaming processing, wireless sensor networks, and wireless protocols) and motivate how it makes progress with regard to the state of the art in terms of a distributed solution. From the other side, the paper focuses on three specific aspects of the middleware while other ones are not addressed. For instance, a relevant issue is how to distribute the computations on the resources. This is a quite challenging aspect, especially when the architecture comprises a large set of nodes and there are many applications to be executed, and multiple applications are allocated on the same node. Nevertheless, sometimes the discussion is quite verbose; more schemes and figures would have helped.

In conclusion, it is a nice and interesting read.