Improvements to incentives in incentive-driven blockchain technologies
- Autores: Conor McMenamin
- Directores de la Tesis: Vanesa Daza Fernández (dir. tes.), Matteo Pontecorvi (codir. tes.)
- Lectura: En la Universitat Pompeu Fabra ( España ) en 2023
- Idioma: español
- Tribunal Calificador de la Tesis: Ittay Eyal (presid.), Juan Pedro Moreno (secret.), Sara Tucci-Piergiovani (voc.)
- Programa de doctorado: Programa de Doctorado en Tecnologías de la Información y las Comunicaciones por la Universidad Pompeu Fabra
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
- català
Aquesta tesis presenta una serie de protocols en àrees que pertanyen al nucli de la tecnologia Blockchain com són els algorismes de consens, les finances descentralitzades o la computació distribuïda. Aquestes àrees estan unides per una dependència crítica dels incentius. Malgrat això, els estàndards dels protocols existents en cadascuna d’aquestes àrees són vulnerables en l’ús que fan dels incentius, fet que limita el seu atractiu en comparació amb les alternatives centralitzades. En aquesta tesi identifiquem primer els defectes dels estàndards existents en aquestes àrees pel que fa als incentius. A continuació, aproximem de manera comuna i genèrica aquests defectes i per a cada escenari, introduim un protocol que evita les mancances detectades als protocols existents.
- English
In this thesis, we present a set of protocols in areas at the core of current blockchain technology literature; consensus, decentralized finance and distributed computing. These areas are bound by a critical dependency on incentivization. Despite this, existing protocol standards in each of these areas are vulnerable to well-documented incentivization exploits which limit their attractiveness compared to centralized alternatives. We first identify the shortcomings of existing standards in these areas with respect to incentivization. We then take a common, general approach to these shortcomings, and in each instance, propose a novel protocol with which to address the incentivization shortcomings we identify.
At the consensus level, this disconnect is highlighted in Fruitchains, one of the most referenced academic works in relation to fairness of reward distribution and incentive compatibility. Fruitchains crucially relies on an underlying blockchain satisfying a state machine replication (SMR) protocol in order to guarantee fairness of rewards on the chain distributing the rewards. The authors fail to consider the incentives of all parts of the system, relying on an altruistic majority of players participating in an underlying consensus protocol to fairly reward all players for the work they have done. The fallacy in this reasoning can be seen through the simple strategy of only rewarding oneself and not rewarding other players in the Fruitchains protocol (not pointing to other players¿ ¿fruits¿). This strategy strictly dominates the Fruitchains strategy with respect to maximizing rewards share. Oversights like this are ingrained at the foundations of distributed-system incentivization research, and it is these oversights which stand as the focus of this thesis. At the core of the issues we identified in consensus-level incentives was the dependency on one or more altruistic, honest-by-default players (see Section 3.2 for an extended summary of these findings). Within the foundational BAR (Byzantine, Altruistic and Rational) player model to which distributed systems literature on incentives typically reference, such a dependency is allowed. This motivates the first contribution of the thesis. In Chapter 3, we propose the ByRa player model, a player model free from altruistic player dependencies within which incentives are required to be more robust. Within this new player model, we propose Tenderstake, a consensus-level protocol which is incentive compatible under a rational majority and adversarial minority of players, and importantly, without any need for honest players. Beyond consensus-level incentives, application-level incentives in blockchain-based systems are not themselves without flaw, and form the basis for the remainder of the thesis. The most prominent research on blockchain-based application-level incentives can almost surely be traced back to the now seminal work of Flashboys 2.0. In Flashboys 2.0, the authors examine the smart-contract enabled Ethereum blockchain at a transaction level. The authors identify that the block-producer privilege of being able to order, censor and create transactions is highly profitable, beyond the base block rewards for creating valid blocks. This profitability was coming at the expense of the users submitting transactions for inclusion in the blockchain, in the now infamous phenomenon of MEV (then miner extractable value, now maximal extractable value as this value extraction is not necessarily exclusive to the miner). Although the incentives in mainstream blockchains were misaligned at the consensus-layer (as demonstrated in Selfish Mining), the price of anarchy for deviating from the prescribed protocol and reducing trust in the underlying blockchain seemed to be ensuring that consensus-level incentives were performing more-or-less as required. However, it is clear that application-level incentives were not. The foundational example for this can be seen in the Uniswap decentralized exchange (DEX) protocol. An automated market-maker, accessible by all, allowing swaps from one token to almost any other was a simple, seemingly effective public good, yet overall limited, and deceptively expensive to interact with. Announcing an intent to trade in an exact size, price and direction before it gets confirmed, as is done in Uniswap and most other DEX implementations, is highly exploitable in a competitive trading environment. The quantified losses sustained by users submitting transactions to Uniswap are in the hundreds of millions of dollars. Exact figures are impossible to quantify, but are bounded below by these recorded losses. The total value extracted from users across all protocols and blockchains is likely therefore to be in the billions of dollars. This can be seen as a much more pressing issue than consensus-level incentives with no glaring ongoing exploits, and one in which research is identified as lacking. At the core of MEV is the informational advantage held by some subset of players in the system when deciding what action to take. If a player can predict with some probability greater than random how the underlying state machine will update (by observing the transactions likely to be accepted in the coming blocks for example), players¿ strategies change dramatically. In the DEX example, a player aware of an imminent buy imbalance for a particular token in upcoming blocks gives that player the ability to front-run the information, and/or back-run the information when this buy imbalance is likely to cease. This issue forms the basis for Chapters 5 and 6. In these chapters we highlight this advantage, and construct protocols under the assumption that this information will be used. In this paradigm, we propose two new DEX variants which protect uninformed users from the negative externalities of such informational disadvantages.
With respect to randomness-based selection, and/or rewarding based on popular choice, key components in distributed computation (DC) outsourcing, the problem is no different. Unfortunately, research in the area is significantly more detached from the realities of incentives, focusing on scalability, speed, and applications, despite computation outsourcing being a billion dollar privatized industry. For decentralized DC to compete with such a behemoth, incentivization is needed. There are a litany of works related to DC incentives which fail to capture a world in which computers must be rewarded financially for computations, instead depending on computer utility to be measured in something directly related to the quality of the computations they are performing. Like using an SMR protocol to incentivize SMR], this is another incentives chicken-and-egg scenario, rendering the potential of DC moot in the real world. In Chapter 4 we address some of the major gaps in these previous works on incentivization in DC, providing a generic protocol for blockchain-based DC which is incentive compatible in the ByRa model. Following on from the foundation on incentives and blockchain formalization provided in Chapter 3, the remainder of the thesis can be seen as a road-map to tackle some of the key open problems in decentralized protocol incentives. This is the backdrop for Chapters 4, 5 and 6 in which we provide incentive-specific improvements on industry standards with regards to distributed computing, decentralized exchange and automated market-makers respectively. The contributions of all chapters can be summarized as follows.
Chapter 3: Tenderstake. We outline the ByRa model, a new player model free of altruistic players, and argue that this player model more accurately describes players in distributed games. We outline the properties of strong incentive compatibility in expectation and fairness as properties that distributed consensus protocols must possess in the ByRa model in order to guarantee state machine replication. We then provide Tenderstake as a protocol which possesses these properties.
Chapter 4: Marvel DC. We present Marvel DC, a blockchain-based strong incentive compatible distributed computing protocol. Marvel DC stands as an improvement on existing industry standards in which computers are altruistic, or trust third parties are required to ensure honest behaviour. We also provide Privacy Marvel DC, a privacy enhancement for Marvel DC which decouples computation outputs from the computers who computed them, making Privacy Marvel DC appropriate for computations in which outputs potentially reveal sensitive information about the computers computing them, such as Federated Learning.
Chapter 5: FairTraDEX. We present FairTraDEX, a decentralized exchange protocol based on frequent batch auctions in which the game-theoretic optimal strategy for all players is to trade at the true market-implied price of the underlying token swap, excluding explicit fees.
Chapter 6: Diamond. We present Diamond, an automated market making protocol that aligns the incentives of liquidity providers and block producers in the protocol-level retention of losses-versus-rebalancing (loss-versus-rebalancing is explained in Section 6.3.2). In Diamond, block producers effectively auction the right to capture any arbitrage that exists between the external market price of a Diamond pool, and the price of the pool itself. The proceeds of these auctions are shared by the Diamond pool and block producer in a way that is proven to remain incentive compatible for the block producer.
- català
