This post was guest-authored by Crypeli and published here with permission.
From the previous article, professionalized mining is playing an increasingly important role in the ETH network. It’s no mystery that they are around, as Hudson Jameson, Core ETH team member, stated: “…at this point ASICs have been out for months.”. Nevertheless, and even though there’s some clarity on the fact that ASICs are around, it may be convenient to gauge how capable actors related to ASICs can perform and what that means for the ETH blockchain.
In order to make such an approximation, estimations on how convenient a purchase of an ASIC would be are performed for 3 key actors: retail miners, professionalized miners and ASIC makers. The first two actors were described in my previous article, but the last one has two (known) entities: Bitmain and Innosilicon. The former has the E3 model, which boasts 190 MH/s at a rating of 760 W and a published price of USD 1262; the latter has its A10 ETHMaster, which comes in 3 flavors: 356, 432, and 485 MH/s with a rating of 650, 740, and 850 W and a published price of $3,800, $4,400, and $5,000, respectively (Source: https://2miners.com/blog/asic-miners-for-ethereum-antminer-e3-vs-innosilicon-a10-eth-master-comparison/).
In order to estimate what the acquisition cost for the aforementioned actors would be, these assumptions were made:
• Retail miners can only acquire ASICS at published (i.e. retail) rate;
• ASIC makers will be able to make their own ASICs at their internal cost. It is almost impossible to know their real cost structure, but an approximation can be made with data from a company in a similar business: Nvidia. As per their latest press release regarding financial statements, their net profit % (net income/revenue) is 35.3% for Q2 Fiscal 2019. Thus, it is plausible to assume that ASIC makers have similar levels of profits (if not lower; I’ll later explain why), and that the cost of making their ASICs is 35.3% lower than the published price;
• Professionalized miners will be somewhere in the middle. Their profit will be fixed to 20%, for this analysis’ sake;
• As for Innosilicon’s A10, only the 485 MH/S flavor will be taken into account.
Let’s go to the analysis then:
Table 1. Current ASIC mining outlook for ETH.
Highlights:
• Innosilicon’s ASIC has better ROI in the worst-case scenario, which would be an interesting fact for retail miners. Nevertheless, ASICs are useful for nothing more than ETH mining, so this is a risky bet, not to mention its hefty price tag;
• Bitmain’s ASIC is better in the best-case scenario, which is very near to the operating conditions of Bitmain and elite professionalized miners;
• As per the red cell indicates, a retail miner should expect to make ROI in even more than 3 years; PoS should be ready before that, so this is a no-no for retail miners;
• On the other hand, Bitmain has a potential of making ROI in less than 7 months. In fact, it could be even less due to:
◦ Bitmain not having to (until recently) compete, as Nvidia does with AMD;
◦ Bitmain having a lighter cost structure than Nvidia, as it does not spend that much in marketing, foreign locations, etc.
• As a consequence, Bitmain’s ability to ramp up their capabilities and their vertical integration (they make ASICs, they mine with them, they sell them – sometimes used ones – and they repair them) is something that has to be taken into account when making economical decisions about the ETH blockchain.
51% attack as seen from a voting bribery perspective.
Now that it’s clear that ASIC mining is a significant factor in ETH’s mining dynamics, and as such they are a good candidate to be part of a 51% attack, the next logical question would be: how likely a 51% attack is now? Such an attack consists of a malicious, majority share of a PoW blockchain’s hashing capacity forcibly making their preferred branch of the blockchain the longest one. That means there are multiple branches on a blockchain and that miners channel their hashing capacity to their preferred branch, in hopes they earn a fixed block reward for the time between blocks. This sounds very analogous to several candidates on a public election system to which voters assign their vote towards their preferred candidate pursuant to public office for a fixed term. What if we treat ETH mining as a voting system (which it somewhat is, indeed), and a 51% attack as a potential vote purchase spree?
By taking a look at the literature about voting bribery, a way of modeling elections fits very good in ETH’s blockchain: the “plurality-weighted-$bribery” model (Faliszewski et al., 2006 and Faliszewski et al., 2009) where voters (miners) would have a bribery cost (e.g. a small percentage above profitability), and their votes (mining capacity) are weighted (i.e. the weights would be the mining capacity). Another interesting modeling way is the “plurality-weighted-negative-bribery”, which is the same as “plurality-weighted-$bribery”, except that the objective is not to vote for one’s candidate but rather not voting for someone else. Now, a big actor such as Bitmain trying to directly influence other miners would be something noticeable and traceable; this leads to recognize the fact that, as long as, e.g., Bitmain, can indirectly dominate the rest of the candidates, it’s not necessary for Bitmain to own or directly influence the voters. In fact, Bitmain can create “dummy” candidates, i.e. other mining power purchasers, so as to disguise and indirectly influence the ETH network.
All of this may seem far-fetched, but there’s a platform that can allow for this: NiceHash. Simply put, you pay in bitcoins for ETH mining power that you can fully dispose of (and we all know Bitmain is BTC mining’s king). NiceHash has been hacked before. You know what I am thinking about. 😊
Now, in order to optimize their budget, they need to know (or have a reasonable estimate of) 1) The mining costs in a given country (energy, mostly), so as to determine the potential cost per megahash (these data are available, as it was presented in my previous article); 2) How much potential hashrate a country has (this can’t be known for sure, but weighting proxies such as population, GDP, energy consumption, etc. can be explored). Such an optimization problem can be stated as follows:
Sets:
• C: countries in the world
Parameters:
• ci : total cost of bribing miners in country i
• si : share of global ETH hashrate of country i
Decision variables:
• xi : binary variable that represents whether country i will be bribed or not (1 if country I is bribed, 0 otherwise)
Objective function:
Constraints:
• Total budget limitation:
• Minimum share of global hashrate required for a 51% attack:
• Binarity of decision variables:
It is obvious that there are no clear-cut answers for the 51% attack probability here, and that data need to be fed into the aforementioned optimization model to provide some insight. Nonetheless, it serves a conceptual purpose, as it sheds some light on how a potential briber could perform its attack, how it will perform it, the amounts of money it can potentially spend and even which countries or regions it will target. Spoiler: it will likely be in a breakeven KWh price-based ascending fashion, i.e. it will start to indirectly bribe the country that has the lowest breakeven point with little more than the breakeven rate, then the next country if this goes fine with the first one, and so on. It’s the same strategy that you follow when backpacking: you have limited carrying capacity, so you start with the most beneficial items such as food and water, and then the least important ones until your backpack’s capacity is maxed out (for the ones who are familiar with linear optimization, both are Knapsack problems).
My intention here is not to provide a definite answer to ETH’s economical problems, but to propose another approach to tackle them, so that the community can decide for the best course of action to take. If interested in collaborating/helping me continue with this work, or if you have some comments, feel free to comment below, contact me or the Omni Analytics team via my twitter account (@Crypeli1) or the Omni Analytics team’s twitter account (@OmniAnalytics), respectively.