Initiatives such as blockchains and decentralized storage networks are pushing for a decentralized Web3 to replace the current architecture. At the core of Web3 are network resource sharing services, which allow anyone to sell spare network capacity in return for rewards. These services require a way to establish trust, as parties are potentially malicious. This can be achieved by reputation systems.In this paper we make the case for using deep reinforcement learning in Web3 reputation calculation. More specifically, we propose a model which allows for decentralized calculation of scores with high personalization for the user.

Traversing NAT's remains a big issue in P2P networks, and many of the previously proposed solutions are incompatible with truly decentralised emerging applications. Such applications need a decentralised NAT traversal solution without trusted centralised servers. In this paper we present a decentralised, relay-based NAT traversal system, where any reachable node is able to assist an unreachable node in NAT traversal. Smart contracts on the Ethereum blockchain are used to ensure fair rewards. Besides financial incentives, a reputation system based on transactions on-chain is used to mitigate against malicious behaviour, and guide peer discovery. Evaluation of our system shows that a combination of historic performance metrics leads to an optimal scoring function, that the system takes little time to reach stability from inception, and that the system is resilient against various attacks. Implementation of the smart contract shows that the cost for participants is manageable. CCS CONCEPTS • Networks → Peer-to-peer networks; • Security and privacy → Distributed systems security.

Many decentralised services have recently emerged on top of blockchain, offering benefits like privacy, and allowing any node in the network to share its resources. In order to be a competitive alternative to their central counterparts, their performance needs to match up. Specifically, service allocation remains a performance bottleneck for many decentralised services.In this paper we present FLOCK, an allocation system which is highly scalable, fast, and lightweight. Furthermore, it allows nodes to indicate their preference for clients/sellers without needing to submit bids by using stable matching algorithms. We decouple the price discovery and outsource this function to a smart contract on the blockchain.Additionally, another smart contract is used to orchestrate the allocation and take care of service discovery, while trusted execution environments securely compute allocation solutions, and off-chain payment networks are used to send rewards.Evaluation of FLOCK shows that gas costs are manageable and improve upon other solutions which leverage auctions, and that our instance of the stable matching algorithm greatly improves run-time and throughput over auction counterparts. Finally, our discussion outlines practical improvements to further increase performance.