Trading Network Performance for Cash in the Bitcoin Blockchain

Abstract

This thesis describes a longitudinal study of Bitcoin,\nthe perhaps most popular blockchain based system today.\nPublic blockchains have emerged as a plausible messaging substrate\nfor applications that require highly reliable communication.\nHowever, sending messages over existing blockchains can be cumbersome\nand costly as miners require payment to establish consensus on the\nsequence of messages, since the electricity consumption\nneeded to run miners is not negligible.\nThe blockchain protocol requires an always\ngrowing size of the information stored in it so its scalability is\nthe biggest problem. For that reason we decided to\ncollect and store data locally in our own data structure,\nnecessary for the analysis,\nallowing us to save up to 10 times the amount of disk space.\nToday, systems using the blockchain protocol are emerging,\nand cryptocurrencies are a glaring example\nof its implementation. Bitcoin\nrepresents the largest cryptocurrency on market,\nand it has to face a massive scale due to its popularity,\nhaving in 2012 about fifty thousands\ntransaction per day and reaching now,\nin 2017, more than three hundred fifty\nthousands of transactions\napproved every day.\n\nThis massive scale in the system leads to a saturation\nof the messaging substrate, hence performance issues.\nIn this thesis we will focus also on the Bitcoin network\nperformance, in particular, transaction throughput and\nlatency.\nFrom 2009 to 2017 a lot of analyses on\nthe blockchain have been performed,\nenhancing the considerable change in\nthe block size limit,\nfrom 256 bytes to 1MB,\nas an attempt to overcome scalability problems.\nDifferent papers were published, discussing\nwhether changing or not the block size limit.\nIn addition, the Bitcoin price increased\nfrom ~0.7$ to more than 7.000$,\nmaking the system even more desirable for\nminers, but causing several complications\nin the fee and reward mechanism.\nWe evaluate and discuss possible ways to improve this fee\nmechanism in order to guarantee more revenue for miners along\nwith an user fee optimization.\n\nWe finally present our own system for\nlongitudinal analysis on the Bitcoin blockchain,\nBAS. It generates a dataset\nwhich contains a significant portion of\nthe whole blockchain, updated on September 2017.\nWe discuss our results and compare them with\nother evaluations from past years, considering\nthree main key points: scalability,\nperformance and fees/costs.\nWe discuss how scalability affects performance,\nand how the costs and fees are dependent\nfrom them both.\nWe want also to take into consideration\nthe environmental impact of Bitcoin\nand how it affects the coming\nof new cryptocurrencies.\nWe evaluate and\npropose, using machine learning techniques,\ntwo different cost prediction models that aim to\npredict bandwidth for upcoming transactions\naccording the fee they are willing to pay, and\nthe expected revenue for miners according to\nthe time spent mining.\nThese models can\nbe used by application to throttle network traffic to optimize\nmessage delivery. We also discuss\nwhether the block size limit should be increased for a higher\nthroughput or not.