Churn in the Bitcoin Network
Muhammad Anas Imtiaz, David Starobinski, Ari Trachtenberg, Nabeel Younis
Abstract
Efficient and reliable propagation of blocks is vital to the scalability of the Bitcoin network. As a result, several schemes, such as the compact block protocol (BIP 152), have been proposed over the last few years to speed up the block propagation. Even so, we provide experimental evidence that (i) the vast majority (97%) of Bitcoin nodes exhibit only intermittent network connectivity ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.</i> , churn), and (ii) this churn results in significant number of unsuccessful compact blocks, roughly three times the statistic for continuously connected nodes. We conduct experiments on the Bitcoin network that show that churn results in a roughly five fold increase in block propagation time ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.</i> , 566.89 ms vs. 109.31 ms) on average. To effect our analysis, we develop a statistical model for churn, based on empirical network data, and use this model to actuate live test nodes on the Bitcoin network. The performance of the system is measured within a novel framework that we developed for logging the internal behavior of a Bitcoin node, and which we share for public use. Finally, to mitigate the problem of missing transactions in churning nodes, we propose and implement into Bitcoin Core a new synchronization protocol, dubbed MempoolSync. Our measurements show that churning nodes implementing MempoolSync experience significantly better performance than standard nodes not implementing MempoolSync, including average block propagation delay reduced by over 50%.