Résilience à la collusion dans les mécanismes de places de marché décentralisées

Decentralized marketplaces in Web3 aim to protect against censorship, bias, and single points of failure that may exist in their centralized counterparts. Still, some mechanisms tend to remain centralized, for example the search mechanism that enables discovery of new assets in the market. Such vulnerabilities have been exploited in live marketplaces in recent years: it is all the more essential to provide protection mechanisms. In this thesis, we propose protocols to uphold the reliability and fairness of marketplace mechanisms, notably through resilience against colluding malicious actors. First, to address decentralized selection of a subset of participants among a population comprising malicious actors, we contribute a blockchain-based protocol to avoid malicious actors swaying selection to their benefit. Then, considering selected sets of participants that will work together on tasks in a decentralized computing marketplace, in an environment with no access to trustworthy or non-confidential monitoring information, we present an incentive mechanism that collectively punishes or rewards task participants based on the outcome of their tasks. We also describe and evaluate how to meet a target success rate for the marketplace's tasks: our proposed algorithm is able to meet such targets and to reduce the failure rate by 5 to 10 times compared to an unprotected system. Additionally, we show how providers of a marketplace's search mechanism can favor a subset of search consumers, granting them an unfair advantage in accessing information about the most recent state of the market. We protect decentralized marketplaces' search with our protocol COoL-TEE, which enables honest search consumers to avoid malicious search providers, who selectively delay responses to benefit colluding consumers. Honest consumers collaborate with Trusted Execution Environments (TEEs) inside the host providers, in order to select close, fast, and honest providers. Using simulations of consumers sending search requests from around the globe to geo-distributed providers hosted in datacenters, we illustrate how COoL-TEE reduces malicious advantage close to a scenario without attacks. Finally, many TEE and traditional protocols rely on trustworthy time measurements for their execution logic, including COoL-TEE. However, attackers controlling the operating system are capable of attacking the TEE's time perception and, in turn, of manipulating the protocols depending on the timestamps. We contribute a public implementation of the state-of-the-art but closed-source protocol Triad and empirically showcase attacks. Calibration can be manipulated to affect the TEE's perceived clock speed. Furthermore, attacks on a compromised machine could propagate to honest machines participating in Triad's trusted time protocol. We discuss mitigations to these vulnerabilities for higher resilience against such attacks.