An Efficient Miner Strategy for Selecting Cryptocurrency Transactions

Saulo dos Santos, Chukwuka Chukwuocha, Shahin Kamali, Ruppa K. Thulasiram

Abstract: Cryptocurrencies like Bitcoin use the blockchain technology to record transactions in a distributed and secure way. Each block contains a cryptographic hash of the previous block in addition to a set of transactions that it records. The first step for a miner to add a new block to the blockchain is to select a set of pending transactions from a mempool. The total size of selected transactions should not exceed the fixed capacity of blocks. If a miner completes the computationally-hard task of finding the cryptographic hash of the formed block, the block can be added to the blockchain in which case the transactions in that block will become complete. Transaction might have a fee that is granted to the miner upon being complete. As such, when forming a new block, miners tend to select transactions that offer the best fees. Finding a set of transactions with maximum total fee that fit into a block translates to the classic knapsack problem, which is an NP-hard problem. Meanwhile, miners are in fierce competition for mining blocks and hence cannot dedicate too much computational power for selecting the best set of transactions. Most existing solutions to tackle this problem are based on sorting the set of pending transactions by the ratio between their fee and their size. While the sorting step is not a bottleneck in normal situations, transaction can grow explosively in case of a market turbulence like that of 2017. Meanwhile, the total number of transactions increases over time. As such, it is desirable to have an efficient strategy that does not rely on sorting transactions before forming a block. In this paper, we review some of the existing strategies for miners to select transactions from the mempool. We also introduce a robust solution called Size-Density Table (SDT) for selecting transactions that does not use sorting. We perform a theoretical and experimental analysis of our solutions to compare it with other strategies. Our results indicate that our algorithm runs faster than previous approaches while the quality of its solutions (the total fees collected in its blocks) is also slightly better than the best existing strategies.