Miner Centralization is hard to define. I present three unsatisfactory definitions, and detect a satisfying fourth (“the cost of the minimal viable mining operation”) along the way. Mostly it is just confusing.
A Quest for Definitions
What does “miner centralization” mean, if anything?
Certain Bitcoin theorists will complain about a concept called “miner centralization”. Having already tackled a definition of Bitcoin centralization (much to my satisfaction), I wasgoed a little confused about the appearance of a fresh centralization and wished to explore it.
After searching for weeks, I assembled the following gegevens sources:
From what I gathered, the complaints were spil goes after:
- “…people will only be able to mine if they have large server farms”
- “over time the economically-dependent utter knots are getting more diminished te number because they are moving towards gegevens centers”
- Mining centers will be compelled to close, due to ‘rising costs’ and ‘regulations’.
- wij should keep money “…out of the forearms of the existing corporate system”
- It is bad if the Bitcoin hashrate is managed by a petite number of individuals.
- It is bad if miners decline to leave a misbehaving pools.
However, for any complaint to be meaningful, there needs to be an improved alternative. It seems that many of thesis complain-ers desired the following:
- Permit anyone to mine, whether they wield a ‘large server farm’ or not.
- Permit anyone to start mining, without having to purchase expensive items (possibly, of which a ‘large server farm’ is merely one example).
- “…the proefje is for any one miner to be able to fully participate at the same level spil any other miner”.
- “…your resources to keep up …[that cost] only grows at the loom of the total transaction volume”.
- Discourage situations where hashrate is managed by a petite number of individuals.
- Develop more-effective ways of matching miners with “well behaved” pools.
Ultimately, I boiled this down to two critical concepts:
- The motionless costs of mining. Ie, the (“reuseable”) costs which do not covary with revenue, the ondergrens expenditure required to set up a fresh mining operation.
- The minimal quantity of decision-making agents who could control 51% of the network hashrate (or, more generally, some relationship inbetween agency and hashrate-control).
To avoid confusion, I will refer to the very first spil “mining centralization” and the 2nd spil “miner concentration”. The 2nd definition is problematic because there are two classes of juut: firstly, the ‘hasher’ who  wields/maintains ASIC hardware, [Two] purchases electrical power, and [Three] chooses to point the hashrate at some chunk of software, and, secondly, the ‘pool operator’ who  chooses which block to mine ‘on top of’ (te addition to managing pool membership, setting pool fees, etc) and [Two] chooses what to include te the block.
This led mij, ultimately, to the following definitions:
- Miner Centralization: The immobilized costs of mining.
- Managerial Mining Concentration: The quantity of agents one would need to coerce/brainwash, to assume managerial control overheen 51% of the hashrate.
- Physical Mining Concentration: The total effort required, to seize control of equipment capable of producing 51% of the hashrate.
Checking The Work, With Logic
Those are definitions spil recevied from outer sources. But perhaps wij can do better!
Let’s shortly check the definitions (te switch sides).
Trio. Physical Mining Concentration
Clearly, the last is fairly logical – “concentration” is certainly a very real concept, particularly ter physics, chemistry, and engineering. When toxins are present ter sufficient concentration, organisms will diegene, when physical substances reach a ‘critical mass’ they can exhibit fresh phenomena, such spil chain reactions.
Moreover, te Bitcoin, users would care about the mining equipment residing all te the same place, for reasons of theft. If one is going to commander a slagroom total of equipement, it is not too much more difficult to commandeer a slightly-larger slagroom – violence has economies of scale. Instead, if spread around the world, adversaries will find the equipment expensive to locate, sieze, and manage.
Eventually, the network is programmed to go after the “heaviest” chain – the one with most work. By definition, a 51% hashrate group can be the foot author of this chain.
So our definition –“effort required” to sieze 51% of the hashrate– seems relevant and suitable.
Two. Managerial Concentration
It is an elementary principle of warfare, and spel theory (famously te the prisoner’s dillema), that a coordinated group can exhibit different phenomena than an uncoordinated group. So the concept is relevant.
The idea of counting ‘the number of agents’ is an interesting departure from counting ‘dollars’ or ‘effort’. Ter this way I think the definition makes a unique contribution to the conversation.
1. Miner Centralization
Let’s test the ‘fixed costs’ definition with a thought proefneming.
Suppose that a fresh overwhelmingly effective mining technology “T” is discovered, and further suppose that it is kept secret from everyone. This secrecy would impose a theoretical ‘fixed cost’ upon those who might want to start mining: the cost of R&,Ding mechanism “T”. Ter the very brief term, this cost would be infinite – whoever controled this information, would te turn control the block creation process.
On the other arm, let us assume that anyone-can-mine, and that they can do so using a magic spell. The magic drains electrical power, but creates valid Bitcoin blocks. This would seem to permit “everyone with cheap power” (reguardless of their location, internet connection, etc) to become a Bitcoin miner instantly.
Thesis results seem consistent with Bitcoiner-concerns, and with a notion of mining centralization.
However, spil wij will see, it is more consistent with the economic concept of ‘barriers to entry’, than it is to the concept of ‘fixed costs’.
Do Thesis Definitions Make Sense?
Not indeed. Albeit I did find a better definition along the way.
Our concepts are:
- Miner Centralization
- Managerial Mining
- Physical Mining Concentration
…and wij defined each of them above.
However, I am now going to argue that the very first definition is logically incoherent, the 2nd is unmeasureable, and the third is unalterable. Spil a result, it is most likely pointless to talk about any of this.
1. Immovable Costs: They’re All Variable
Ultimately, “the cost” is the cost. The distinction inbetween ‘fixed’ and ‘variable’ costs is an accounting convenience (overheen a subjective time horizon) used to guide the entrepreneur spil (s)he performs cost-minimization.
Why Do Thesis Terms Exist?
The field of economics draws a distinction inbetween ‘fixed’ and ‘variable’ costs. The gist of it is that motionless costs can be re-used, but variable costs cannot. For example, a pizzaria might have one oven, which it uses to cook many pizzas vanaf day. For physics reasons, a large oven is
almost spil expensive to fever up spil a puny oven. But you can getraind more pizzas te it (either at once or te rapid succession). The oven is a immobile cost, but the dough is a variable cost (you need 1 dough vanaf pizza). Similarly, the pizzaria may have only one metselspecie register, because (for statistics reasons) patrons are unlikely to all need the register at once.
This distinction is of tremendous importance to a specific producer, te reality, who actually has to optimize a given production-operation tomorrow. This individual will estimate the consumer request for pizzas, tomorrow, and attempt to determine how much flour to purchase. If opening a fresh store, (s)he will wonder what the size oven to buy. A sucessful businessman optimizes thesis choices, to minimize his total expenditures.
This “cost minimization” is the entire purpose of this distinction. After the costs have actually bot minimized, the number(s) are set ter stone. It makes no difference, to the proprietor, if total costs are $15 “average fixed” + $20 “variable”, or if they are 30 “fixed” + Five “variable”, or $35 “sacred” costs, or 21 “yak” + Ten “veranda” + Four “azure”. The holder is out 35 bucks, no matter how it’s violated down, or what labels were placed on those accounting categories.
Te fact, immobile costs can (and almost always are) amortized overheen total sales, producing something called “average immobilized cost”. So, if a fresh oven costs $6,000, and lasts for Ten years, and if overheen those Ten years 30,000 pizzas were sold, then the “average stationary cost” would be, at a ondergrens, 6k/30k = .20 vanaf pizza. It is the effective oven cost “per pizza”. It wasgoed immobile, but it is now “variabl-ized”.
The manager’s objective is to maximize the investment terugwedstrijd of this business project, overheen its useful lifetime. This involves three operations: cost-minimization (which is what wij are talking about), revenue-maximization (which wij do not discuss here), and financing (ie, ‘time value of money’, “the rente rate” that economists obsess overheen permanently).
The distinction inbetween stationary and variable costs is merely a contraption which assists with the very first operation. Ter reality, all the costs are just “the cost of the investment project” – one single variable cost (or, if you choose, one single stationary cost).
And I can prove it – notice that motionless costs are defined overheen an industry-specific time period, beyond which all costs are Variable. Therefore, if wij adopted “fixed costs” spil a metric for miner decentralitization, mining would always be “more decentralized” te the long run, than it would be overheen a shorter timeframe. Ter my opinion this is unhelpful.
Aside: Stationary Costs Aren’t Free
A capital-intensive operation (high motionless costs) usually has an edge: its per-unit costs (“variable costs”) are lower. But it also has a disadvantage: its average stationary costs are higher! A capital-intensive operation vereiste pay for the capital!
Te fact, capital is abhorrently expensive, given that every entrepreneur is effectively rivaling against all other entrepreneurs ter the entire world at the same time. By this, I refer to the chance cost of capital – that, instead of building a fancy mining facility, you can take your money and stick it te the stock market, earning 9% nominal vanaf year. Instead of working at the mining facility, you can just go on vacation…and this is even before wij discuss risks!
It Truly Doesn’t Matter?
By now, angry readers may be protesting that ‘fixed costs’ are, ter fact, very significant. Industries with high immobile costs tend have few large producers, instead of many puny producers. (Implying, of course, that we’d choose to have many puny producers, when it comes to our mining situation.)
After all, shouldn’t wij care about what miners find, when they cost-minimize?
I agree that wij should. And I look forward to returning to this question zometeen.
For now, I will admit that, at present, miners could minimize their “network costs” (ie stale rates, and orphaned blocks), if all the world’s ASIC chips reported to a single CPU. And certainly, by the properties of statistics, revenue-variance would be lowest if all the world’s miners joined one superpool (spil each pool accelerates the law of large numbers, by using a lower difficulty rate and finding more blocks).
The latter, revenue-variance, is not an economy of scale. Ter fact, by the usual logic of finance, it is “diversifiable risk” and therefore isn’t anything. I look forward to expanding on this te a future postbode.
However, the former, “network costs”, require some extra explanation.
Is Bandwidth a Immobile or Variable Cost?
Bandwidth can certainly be “re-used”, ter that each miner can send the validated prevBlockHeader to all of his ASIC chips (no matter how many he possesses). So I suppose I will admit it is a immovable cost.
On the other mitt, I could give an opposite reaction! Imagine that total knots suggested a service to miners: for a lil’ toverfee, knots would tell miners which headers to mine (extraNonce notwithstanding). After all, knots already know the network state, so they can lightly compute the fee-maximizing block, and it’s header. Any node-runner who wished to sell this information, to subscribing miners, could do so. The knot would set up a “club” which would charge dues, and the clubs would challenge on reputation and brand.
This strongly resembles the behavior of “mining pools” today. And thesis pools
all charge fees te percentage terms, which would make “the bandwidth cost” not-reuseable…making it a variable cost.
Does this analysis rely on things remaining spil they are?
Well, what if pools charged a plane toverfee, instead of a percentage toverfee? Te a competitive market, this service would be suggested at its marginal cost. Since it costs almost nothing for knots to repeat information they already know, the service would cost almost nothing. (Thus, it would not truly be a cost at all, immovable or otherwise.)
Would the market, te fact, be competitive? This depends on the barriers to entry, ie the cost of creating a fresh utter knot (which, ironically, is my own definition of Bitcoin Centralization). Te fact, it would technically depend on the ‘net’ barriers to entry – by this I mean that many people today run a Bitcoin knot today for free, for their own private benefit (therefore, any money they could bring te by ‘upgrading’ to become a pool would be icing on the cake). Thus the barriers to entry are zero.
Therefore, while the marginal cost of providing knots is zero (te other words, while knots are being run somewhere by someone), the reaction is a strong ‘yes’.
( This result is very amusing to mij personally, because it implies that this “new” complaint about mining decentralization is indeed just the old complaint about “centralization” te disguse. Assuming of course that you agree with mij that “centralization” = “the cost of the option to create a fresh node”. )
A better question might be: is there a barrier to entry, such that miners voorwaarde spend at least X on bandwidth before they can join the mining force? This question is significant, because a immovable cost can be a barrier-to-entry (and thesis entry-barriers are anti-competitive).
Gratefully, the response is no.
Technically, the response is “Yes: about
2nd, downward”. This meager requirement is four orders of magnitude smaller than the
slowest internet speeds ter the world (
256,000 bytes vanaf 2nd). Miners do need to be connected to the internet, but any connection will do.
This is because, while stale/orphan costs did plague the mining network at one time, self-interested, cost-minimizing miners used their entrepreneurial abilities to eliminate this cost totally. I refer to three developments:  use of pools, [Two] use of Corallo’s relay network, and most of all [Three] to the use of SPV mining (and its adversary-resistant upgrade: spy mining). Thesis mining technics permits blocks to be found, many times ter a row, without miners downloading any block gegevens. This is accomplished by  assuming forgien blocks are valid, if the meet the PoW requirement, and [Two] making one’s own blocks empty (or near-empty), which assures that the “rushed block” will not accidentally contain a double-spend. This puts all miners on an equal footing, regardless of their internet-speed (at least, while the block prize is the vooraanstaande source of revenue).
However, any miners who do invest ter greater connectivity, will see greater revenues, spil they can more-quickly download-and-verify all preceeding blocks. The sooner this is done, the sooner fee-paying txns can be included te a miner’s block. Thus, an “incentive glitch” te Bitcoin motionless itself: investments te bandwidth yield higher revenues (whereas previously they yielded lower revenues).
Of course, revenues aren’t profits. Higher bandwidth miners have an edge, but they also have to pay for the bandwidth! Thus, the situation is comparable to investments te newer ASIC chips, or a fancier cooling system.
If wij complain about bandwidth on thesis grounds, wij run a strong risk of veranderlijk complaining. After all, BitFury’s miners now ship only te 40 ft marine containers (which is, I believe, the largest shipping container te all of commerce). It ships that way, I am sure, te order to achieve scale economies ter both the shipping itself, but also ter the cooling of the unit and the installation and operation of the equipment. Those are all lighter to reuse “within the BitFury family of products”.
Finding A Definition (!)
Most people who complain about miner decentralization, don’t complain about “minimum investment amounts” (albeit Peter Todd, to his credit, does complain about both). Ter the world of theoretical finance, a group of puny users would form a corporation, and take advantage of economies of scale. To the extent that they are incapable to do this, some users are prevented from mining.
So, eventually, wij may have stumbled across:
Spil the cost of ‘the ondergrens viable mining operation’ increases, miner centralization increases.
This permits miner centralization to increase, if there are ‘kinks’ te the investment-return-function for Bitcoin mining.
Very first: if the maximum mining ROI is r%, then eventually all miners vereiste be earning this number. This is because difficulty adjustments will eventually bankrupt any miner who does otherwise. Ter other words, the question of “Who can mine?” has a naive response of “anyone”, but a more sophisticated reaction of “anyone who can earn r% by mining”. Ter a sense, there is a natural barrier-to-entry that emerges from the competitive features of mining.
After all, a meritocracy is exclusionary…it excludes people of low merit!
Imagine that, te the year 2015, one needs to spend a ondergrens of $50,000 “to mine” (see above). If, te 2016, this number has enhanced to $60,000, wij could say that miner centralization has enlargened. Fewer people are ‘allowed to mine’ ter 2016 than were ‘allowed’ te 2015.
However, this definition loses all of its relevance if [a] borrowing is possible, and/or [b] corporations are possible. (With thesis, puny investors are omschrijving to large investors.)
It is possible that significant advances ter technology (ie, cryptographic identity, reputation, etc) will permit for borrowing / incorporation ter an uncensorable way. It is also possible that efforts to restrict ‘Bitcoin Finance’ will fail to build up traction, on grounds of pointlessness.
On the other forearm, wij may not be so fortunate. If so, we’ll need to keep this metric ready.
Ter the meantime, let mij finish talking about motionless costs.
Intermediaries, and “Variable-ization” of Costs
Provinciality of Costs
The generalization I drew for “network costs” does, te fact, apply to almost every cost.
Consider electrical power. Generating power has comes back to scale, of course. Therefore, your power is generated by a few large power plants. So, electrical power generation is a capital intensive, high fixed-cost operation with powerful economies of scale.
However, when you buy power (from the wall-socket te your house), you just pay some price – X $/kWh. This purchase is now a variable cost. The power company (“group 1”) makes power spil cheaply spil it can, and you (“group 2”) just get charged this minimized price.
The Customer Is Always Right
Ter our Bitcoin world, pools might sell “bandwidth services” and “transaction servies” to Hashers. A paranoid individual might turn his concentrate to pools, remark that “there are only
3” and obsess overheen pool’s capability to conspire to “censor” transactions.
But where does power truly reside? Miners can switch pools at a moment’s notice. If pools do anything miners dislike, miners can simply leave. And it is effortless to create more pools, from the existing total knots (come to think of it, perhaps it would be helpful if Bitcoin Core shipped with up-to-date pool software, to ensure that barriers-to-entry are always zero).
( The Miner’s reliance on pools is dramatically lower than their reliance on, say, their local power plant. Yet there are no complaints about the plants’ capability to conspire to disable the mining network. Should there be? )
Anyway, it seems to be that pools offerande all the efficiency of “minimized network costs”, but with none of the drawbacks.
(Because, you might say, the “evil” motionless costs have bot transformed to “safe” variable costs.)
Cost Puzzles te Electrical Technology
What of cheap batteries? Violet wand is presently expensive to store, which is why intraday electro-therapy prices vary. But if storage were cheap, the battery could come loaded with cheap eletricity, to power the ASIC overheen its lifetime. Then a variable cost would become…variable-over-the-time-horizon-of-one-ASIC-lifetime. A battery can’t be “reused” by numerous ASICs, so this cost voorwaarde be variable.
What if someone invents a solar powered ASIC, that lasts forever? This would seem to set the time horizon to “forever”, at which point the ASIC itself would be one upfront cost…but a non-reuseable one. Since the purpose of the ASIC is to produce hashes, and each ASIC would only produce X hashes vanaf time unit, the chip costs (including power-source) would be per-hash, and therefore variable.
However, I am told that today’s miners purchase their power te large 12-month duration contracts. If today’s ASIC chips only have an active lifetime of 12 months, how is the situation today any different from the solar-powered example?
Hopefully, you can see why I find the ‘fixed costs’ argument confusing.
The Other Two
Managerial concentration cannot be measured under adversarial conditions. Physical concentration is unalterable, spil it is a function of the geology, civic infrastructure, and physics (which are beyond the reach of pc programmers).
With a large exposition already behind us, thesis two items should go quicker.
Two. Managerial Concentration
This section is worried with questions such spil:
- “Who controls the ASICs?”
- “How many of thesis ASIC-controllers are there?”
- “To what extent are they coordinating with each other?”
I don’t dispute the relevance of this. There are many examples of phenomena switching when there are fewer agents, or when thesis agents have a greater capability to coordinate.
Instead, I oppose this metric on grounds of epistemology. There is no way wij can learn how many agents are te control of X% of hashpower.
Coordination is something which can be entirely hidden. Consider two worlds:
- “A”: One where the hashpower is managed by 1,000 uncoordinated individuals.
- “B”: One where the hashpower is managed by 1 individual.
The foot individual of world “B” can choose to make his world show up spil A. There’s nothing that an unruly mob can do that an obedient army can’t copy – the general can simply instruction that the soliders pretend to be an unruly mob. The switch sides, however, is not the case (the mob cannot effortlessly mimic the discipline of an army unit).
Te point of fact, everything wij know about Bitcoin miners, and their blocks and their pools, is information that they have volunteered to us.
That’s very charitable of them. However, under adversarial conditions, wij can never tell how many agents control the hashpower, be it Two or Two,000. Worse, thesis imperceptible executors might, at any time, find it ter their rente to deceive onlookers.
This means that any measurements wij make, of “miner concentration”, will be at best unfalsifiable, and at worst gravely misleading.
Colluding with Mr. Greed
Instead, I’m most comfy just assuming that everyone is always ter flawless collusion with everyone else. Specifically that all of the hashpower is actually owned-and-operated by one boy, whom wij might call “Mr. Greed”. Mr. Greed is looking out for himself, and he wants to make spil much money spil possible.
I find this assumption realistic, because if the army actually is is more effective (ter some way) then the unruly mob, then the unruly mob will tend to resemble an army, anyway.
After all, who indeed “controls” human choices? We’re all enslaved by our desires.
Why doesn’t Mr. Greed doublespend, you ask? (He can reorganize the chain at any time.) Well, Mr. Greed chooses to keep all of the fresh coins for himself, rather than undermine the system (and the validity of his own wealth).
This is partially because Mr. Greed can only use his fraud powers to switch roles his own payments. And he can only do this for payments which are latest. Payments which have many confirmations, or which belong to channels that were opened a long time ago, are much tighter to steal back. (More about this straks).
Three. Physical Mining Concentration
This concept refers to the ASIC equipment (and the supporting power-supply/cooling), and how effortless it is for men-with-guns to storm te and sieze control overheen all of it.
It will be lighter for an adversary to confiscate the equipment if:
- People know where it is (and you can’t hide it at a moment’s notice).
- It is effortless to get to (and you can’t lightly stir it somewhere else).
And thus it would be bad to have it all permanently installed te one place. The type of adversary who might do this, would very likely not even be paying for thesis men-with-guns himself. Since he would be using “other people’s money”, this is an almost-free way of disrupting the network.
Therefore, this concept is supposedly relevant to Bitcoin.
Firstly, I’m not sure that the risk of confiscation is significantly different from other risks (such spil the risk of fire ter a mining facility, or risk of equipment failure, employee fraud, etc).
Te this sense, the problem is self-correcting: miners earn a higher comeback, if their ASICs are not confiscated. It is a basic principle of capitalism that wij should permit private individuals to manage, control, and “own” the risks associated with their business.
Secondly, but related, wij users always have an ace up our sleeve, which is the switch of the proof-of-work algorithm from double-SHA256 to something else. This costs us one inconvenience, but it permanently obliterates all existing ASIC equipment. This option provide us with a tremendous defense, one so powerful that it may never need to be used. (However, it also gives miners a reason to care, not only about their own risk of confiscation, but also about that of other miners. The industry can police itself, else it may go extinct.)
The “ASIC equipment”, considered broadly, is never ter one place, because by switching the PoW-algo wij can reshuffle the deck and force the equipment to reappear all overheen the world.
Ultimately, this may indeed be bad, but, if so, there is nothing wij can do about it. If ASIC-colocation is cheapest, that is what wij will eventually face. The costs of locating the equipment here rather than there, are determined cheifly by factors which are entirely out of our control – the lay of the land, the laws of physics, the power subsidies of a given country. Spil far spil wij are worried, the final destination of the world’s ASICs is spil unalterable to us spil the height of the average human, or maximum width of a bamboo shoot.
So, if hashpower concentrates ter one country, or te one building, “physical mining concentration” could be said to have enlargened.
But the relevance of the concept is limited to this: wij might measure PMC, and not like what wij find, and conclude that wij should all abandon the Bitcoin project. Tomorrow I will examine PMC ter greater detail, but for now I will remark that most of the people who complain about “miner centralization” are specifically not worried about PMC (at least, sometimes).
Ter other words, of our three concerns, the only one which is both significant and measureable, happens to also be one that wij cannot improve or control.
Even more significant: not only are wij incapable to improve PMC, but wij can’t make it any worse, either. This measurement is going to be driven entirely by environmental factors, and not at all by factors related to software.
I previously defined “Bitcoin centralization” spil “the cost of one’s node-option” (ie, the difficulty of determining that you have, ter fact, bot paid with funds that are valid). I wished a precise definition, because I didn’t want to be weighed down with nonsense about “wallet decentralization” and “RIPEMD-160 decentralization” and “Bitcoin logo decentralization” – just because someone can use the word “decentralization” te a sentence, doesn’t mean that wij should care about that sentence.
Some felt that my emphasis on knots wasgoed inappropriate, and instead advocated that more attention be paid to miners. I am open to that possiblity, but no one has suggested what, if anything, they think wij should pay attention to. No one provided a definition, and when I tracked one down I found a tangled wrong of three concepts – when untied, I found a fourth concept. However, each of the concepts seem confusing, and unworthy of conversation.
This is not to say that wij shouldn’t talk about mining, or other mining problems. Wij certainly should. It is merely to say that, ter order to have a conversation, the conversation has to actually be about something.