Submitted: 11 February 2022
Reject: 26 March 2022
Review 1: 3: (strong accept)
Review 2: -2: (reject)
Review 3: -2: (reject)
Review 4: 0: (borderline paper)
3 (strong accept)
Summary:
The paper presents MobBFT a middleware supporting client-centric state BFT for web applications. The application domain is (small) community driven decentralised networks, such as local markets/fairs where for example one wants to have a common fidelity card. The authors first present their alorithtm to cope with state replication and then they propose their architecture/middleware. An evaluation section validates the proposed approach.
Overall:
The paper is genuinely written and goes well. The idea is pretty neat and is in line with DAIS's topics. The main algorithm is well explained, and is supported with an example which makes it more understandable. Also the overall architecture of the middleware is well argued, and all the key decisions (such as using gossiping protocol for the replica spreading, the use of threes) are well explained. The evaluation phase is conducted in a rigorous way explaining the setting of the experiments and all the envisaged scenarios, which seems credible. In the end, the paper presents a very nice idea/protocol/architecture, which for the proposed application domain, offers nice performances and a valid alternative to the use of blockchains.
-2 (reject)
The paper proposes: A novel specification of byzantine agreement; an algorithm to solve it without explicit message passing and with gossip-based replication of CRDTs; a middleware architecture for browser-based applications; a case study with a shared loyalty program; and an experimental comparison with BFT-Smart and Tendermint protocols. This is an interesting topic and the approach is intriguing.
The paper has however some shortcomings. First, the specification in Section 2 determines that is some value is proposed by a correct replica in view v, then every replica will be able to construct a quorum certificate at view v. This contradicts the assertion that there is no need to store the full log. Having a minimal correctness argument within the paper would be desirable, making it self contained.
On the other hand, the assertion that storage is bounded, in contrast to traditional message-based BFT consensus protocols, needs to be discussed considering that the Removed set in the ORSet CRDT only grows and that traditional protocols have the provision for a snapshot mechanism that truncates the log.
It is also unclear what are the assumptions on the gossip layer and to what extend the CRDT replication ensures them. Namely, is it possible that byzantine nodes eclipse a correct node by systematically avoiding retransmission or manipulating the removed set.
The paper also states that the membership is fixed. Even if the paper does not include a detailed description of a reconfiguration mechanism, it should discuss a general approach of how it is possible. Not having a membership reconfiguration severely hurts the usefulness of the proposal as a middleware architecture.
-2 (reject)
The paper addresses the problem of defining a Byzantine Fault Tolerant (BFT) consensus protocol for client-centric applications.
The paper is motivated by the recent advent of edge computing where many computations are carried on at the client side over devices with limited capabilities in terms of computation power and memory space.
The main contribution is a consensus protocol and its implementation in a browser-based architecture.
The topic is interesting but I have many concerns with the writing and the presentation that prevents me from assessing the correctness of the proposed approach.
In particular,
0 (borderline paper)
The paper presents MobBFT, a peer-to-peer data synchronization framework for decentralized web applications between mistrusting parties. MobBFT combines the benefits of an efficient operation and lightweight setup of a peer-to-peer data synchronization framework with the resilience and fault tolerance of a BFT consensus protocol.
The advantage of the BFT protocol is that it is optimized for unstable network conditions, does not require that all replicas are connected to each other, does not rely on a leader, removing the need for a leader-election procedure which is costly especially when the leader is malicious or loses its network connection temporary.
The problem addressed in the paper is both timely and interesting. The authors do a good job in describing the proposed approach. The paper is written and structured well.
The authors propose an optimistic, leaderless consensus protocol, tolerating Byzantine replicas, combined with a robust and efficient state-based synchronization protocol.
One issue I have is related to the verification of the signatures which is delayed until a decision needs to be made, such as starting a new round or starting a new quorum certificate. This procedure can be quite costly especially in large groups, geo-distributed systems or mobile networks, when many replicas are dishonest. As the authors show experimentally (fig 4b) their protocol does not work very well with large replica sizes resulting in high latencies. Also I do not see clearly the benefit of MobBFT over Tendermint that uses a gossip based approach and rotates the leader often, especially in realistic scenarios.