Uma análise de correlação entre tempo de espera e transações no Ethereum

Abstract

Started in July 2015, Ethereum is the second-largest cryptocurrency in terms of transaction volume at the moment, behind only Bitcoin, which is the pioneer cryptocurrency. Ethereum’s popularity is on the rise, attracting the scientific community. An important issue under discussion regarding the behavior of the platform concerns the mandatory collection of the fee per transaction on Ethereum and its possible relationship with the pending time for confirmation of the transaction. When submitting their transactions to other users, they are required to provide a kind of tip, the mining fee, for their transaction to be included in the network. The common belief in the Ethereum network is that the higher the rate, the shorter the confirmation time. In the present work, a detailed investigation and analysis of this issue were carried out. To conduct this work, 7.2 million transactions were collected on Ethereum. Next, several analyzes were performed on the fees provided by users and the pending time. Firstly, it is verified if there is any correlation between the transaction pending time to enter the network and several resources related to the fees paid to the operators of the Ethereum network for the execution of transactions, through Pearson and Spearman correlations. Subsequently, analyzes grouping transactions by characteristics are conducted in two ways: (i) by absolute values and (ii) by groups defined using clustering algorithms. These groups seek to find users with similar behaviors, trying to find a correlation between the pending time in these cases. Empirical analyzes showed strong evidence that there is no clear correlation between the fees paid by users and the transaction pending time, concluding that such fees, as well as the system characteristics derived from them, inherent to the Ethereum platform, do not determine the time waiting. In addition to assessing the relationship between confirmation time and fee, in this work, a denial of service attack was assessed, a problem that can also affect the confirmation time of transactions on the Ethereum network. Initially, an attack similar to that of distributed denial of service (DDoS) was defined, in which malicious users with the intention of causing slowness in the network insert specific transactions in the network, containing low gas price, increasing the pending time for transactions. To study this attack, here called Penny Attack, an architecture was defined capable of replicating Ethereum’s ETH network in a local structure. Then, simulations were performed to check the system’s response in normal situations of the network, in periods of attack, and also in periods of high network demand. Finally, seven machine learning techniques were proposed seeking better metrics for classifiers capable of identifying the attack proposed here, where methods with accuracy greater than 0.95 and AUC of 0.89 were obtained, providing a reliable model for detecting this attack.