Searcharxiv⌕ Search

arXiv subjects

Houwu Chen

Publications and source records attributed to Houwu Chen.

2 recordsLinked to original sources
arXiv2017

Sky: Opinion Dynamics Based Consensus for P2P Network with Trust Relationships

Traditional Byzantine consensus does not work in P2P network due to Sybil attack while the most prevalent Sybil-proof consensus at present still can't resist adversary with dominant compute power. This paper proposed opinion dynamics based consensus for P2P network with trust relationships, consisting of the sky framework and the sky model. With the sky framework, opinion dynamics can be applied in P2P network for consensus which is Sybil-proof through trust relationships and emerges from local interactions of each node with its direct contacts without topology, global information or even sample of the network involved. The sky model has better performance of convergence than existing models including MR, voter and Sznajd, and its lower bound of fault tolerance performance is also analyzed and proved. Simulations show that our approach can tolerant failures by at least 13% random nodes or 2% top influential nodes while over 96% correct nodes still make correct decision within 70 seconds on the SNAP Wikipedia who-votes-on-whom network for initial configuration of convergence>0.5 with reasonable latencies. Comparing to compute power based consensus, our approach can resist any faulty or malicious nodes by unfollowing them. To the best of our knowledge, it's the first work to bring opinion dynamics to P2P network for consensus.

cs.DC↗
arXiv2017

SkyHash: a Hash Opinion Dynamics Model

This paper proposes the first hash opinion dynamics model, named SkyHash, that can help a P2P network quickly reach consensus on hash opinion. The model consists of a bit layer and a hash layer, each time when a node shapes its new opinion, the bit layer is to determine each bit of a pseudo hash, and the hash layer is to choose a hash opinion with minimum Hamming distance to the pseudo hash. With simulations, we conducted a comprehensive study on the convergence speed of the model by taking into account impacts of various configurations such as network size, node degree, hash size, and initial hash density. Evaluation demonstrates that using our model, consensus can be quickly reached even in large networks. We also developed a denial-of-service (DoS) proof extension for our model. Experiments on the SNAP dataset of the Wikipedia who-votes-on-whom network demonstrate that besides the ability to refuse known ill-behaved nodes, the DoS-proof extended model also outperforms Bitcoin by producing consensus in 45 seconds, and tolerating DoS attack committed by up to 0.9% top influential nodes.

cs.DC↗