What is the discussion about blockchain

A look at current developments in blockchains and their effects on energy consumption

Summary

The enormous power consumption of Bitcoin has meant that in science and practice there are often rather undifferentiated discussions about the sustainability of blockchain and distributed ledger technology in general. However, blockchain technology is already anything but homogeneous - not only with regard to its applications, which now extend far beyond crypto currencies in the economy and the public sector, but also with regard to its technical characteristics and especially its power consumption. This article summarizes the status quo of the power consumption of various implementations of blockchain technology and particularly addresses the recent Bitcoin halving and so-called ZK rollups. We argue that Bitcoin and other proof-of-work blockchains do indeed consume a lot of electricity, but alternative blockchain solutions with significantly lower power consumption are already available today and other promising concepts are being tested that specifically reduce the power consumption of large blockchains Networks could be significantly reduced again in the near future. From this we conclude that the criticism of the power consumption of Bitcoin is legitimate, but that an energy problem for blockchain technology in general should not be derived from it. In many cases in which processes can be digitized or improved with the help of more energy-efficient blockchain variants, energy savings can even be expected on balance.

Abstract

The enormous power consumption of Bitcoin has led to undifferentiated discussions in science and practice about the sustainability of blockchain and distributed ledger technology in general. However, blockchain technology is far from homogeneous — not only with regard to its applications, which now go far beyond cryptocurrencies and have reached businesses and the public sector, but also with regard to its technical characteristics and, in particular, its power consumption. This paper summarizes the status quo of the power consumption of various implementations of blockchain technology, with special emphasis on the recent ‘‘ Bitcoin Halving ’’ and so-called ‘‘ zk-rollups ’’. We argue that although Bitcoin and other proof-of-work blockchains do indeed consume a lot of power, alternative blockchain solutions with significantly lower power consumption are already available today, and new promising concepts are being tested that could further reduce in particulary the power consumption of large blockchain networks in the near future. From this we conclude that although the criticism of Bitcoin’s power consumption is legitimate, it should not be used to derive an energy problem of blockchain technology in general. In many cases in which processes can be digitized or improved with the help of more energy-efficient blockchain variants, one can even expect net energy savings.

introduction

In leading German print media, one can encounter statements such as "The Bitcoin system now uses about as much electricity as the Federal Republic of Germany, and the trend is rising" (Frankfurter Allgemeine Zeitung from 06/06/2020 [1]). On the other hand, 2018 was in the magazine Nature Climate Change published an article according to which, if Bitcoin were adopted on a large scale, the emissions caused by it alone could lead to global warming of more than 2 ° C in the next 3 decades [2]. The FAZ article was modified on our initiative shortly after its publication in the online version and the Nature article was followed by a controversial scientific discussion about the meaningfulness of the underlying assumptions. Nevertheless, such publications lead to an incorrect impression in the public about the ecological consequences of Bitcoin and to an even more problematic generalization to blockchains.

In essence, the statement that Bitcoin and many other cryptocurrencies cause enormous power consumption is correct and important and has been published in numerous publications, including magazines Joules [3,4,5] and Nature Sustainability [6], discussed in detail and justified. Often, however, the striking statements remain present, are taken out of context, inappropriately generalized or used for lines of argument that testify to a lack of understanding of the fundamental relationships between the high power consumption of some cryptocurrencies and economic and technical parameters. For example, the power consumption of Bitcoin neither inevitably increases steadily nor does it grow significantly with the number of transactions processed per unit of time. In addition, blockchain technology is so often mentioned in the same breath as Bitcoin, both in public reporting and partly in science, and is used to explain how blockchains work, that certain prejudices about the power consumption of blockchain technology could generally be established from it .

In fact, there are now numerous cryptocurrencies that are based on technically significantly changed blockchain variants with completely different characteristics in terms of their power consumption. The situation is similar with a large number of implementations of blockchain-based platforms for cross-organizational processes in business and the public sector. In Germany, for example, projects by automobile manufacturers in the supply chain area [7] or the Federal Office for Migration and Refugees [8] should be mentioned. Since the topic of sustainability is rightly very present in politics and business [9], the question of power consumption and the sustainability of blockchain technology in general is very often asked in connection with blockchain-related projects for the reasons described above. The presence of the electricity consumption stigma could therefore significantly impede the adoption of blockchain technology and thus the use of its advantages [10].

Accordingly, in this article we want to give a comprehensive overview of the power consumption of blockchain technology in general in order to create a solid basis for discussion for the general discourse. To this end, we first describe well-known estimates for the energy consumption of Bitcoin, but expand them with a detailed discussion of the recent Bitcoin halving, which reveals many of the basic relationships. With Bitcoin Halving, the number of newly created Bitcoins per block and serving as a reward for the miner is periodically halved every 4 years. This ensures that the number of existing Bitcoins remains limited (geometric series). The aim of this construction is to reduce inflation. On the other hand, we are also examining a significantly larger part of the very heterogeneous spectrum of blockchains as described than just a few cryptocurrencies that are technically closely related to Bitcoin. With that we expand one of us in the magazine Business & Information Systems Engineering published article [11] on the energy consumption of blockchains, which already discusses some of the points addressed in this article and focuses more on the sustainability discussion about cryptocurrency applications of blockchain technology. In comparison, we will go into much more detail here on some aspects that were only briefly discussed there; in particular, in addition to Bitcoin halving, we quantitatively analyze the implications of the use of so-called ZK rollups on the power consumption of blockchains.

Despite the fact that blockchain technology is used far more widely than in Bitcoin and cryptocurrencies, Bitcoin also occupies a central position in this article. This is due to its problematic high energy requirement. We believe that other uses of blockchain technology are far more meaningful.

Basics of Bitcoin and Blockchain Technology

The Bitcoin blockchain was developed to create a decentralized, electronic currency system. In contrast to the transfer of information, it is not easy to transfer (asset) values ​​electronically bilaterally, since electronic objects can be copied as often as required without any effort. Therefore, the information contained in the electronic objects can be valuable in itself, but no value is transferred when the electronic object is handed over or saved [12]. Accordingly, for the electronic transfer of values ​​within a certain group, a so-called register (ledger) recognized by all members is required, in which the ownership structure is entered. A change in this ownership structure in an electronic register can therefore be understood as an electronic value transfer or transaction.

Traditionally, such registers in the form of databases are always kept by trustworthy third parties, such as banks in a currency context. The cryptocurrency Bitcoin [13] presented in a white paper in 2008, which was subsequently implemented and also went into operation in 2009, is based on the decentralized management of the corresponding electronic register through redundant and synchronized (“physically decentralized, logically central”) keeping of the register on all participating computers (nodes). As a result, the validity and execution of transactions is no longer decided by a central authority, but by all participants in the Bitcoin network. The authentication, for example to prove ownership of units of the cryptocurrency or the authorization of payments through this account, takes place with the help of a public key infrastructure and corresponding digital signatures. For the purpose of finding a majority for decisions, the Bitcoin network needs a so-called consensus mechanism, according to which the nodes decide which new transactions are to be recorded in which order.

In principle, such replicated state machines, which guarantee safety and functionality even in the presence of system failures or Byzantine errors, have been intensively researched since 1982 [14] and then implemented in practice with Paxos [15] and PBFT [15] and PBFT [16]. Consensus could be found based on voting, true to the motto “one node, one vote”. What is new with Bitcoin, however, is that not only pre-defined nodes can participate in the network and, accordingly, in finding consensus, but anyone who wants to. This is known as an open, non-restricted system. The election-based process just described is not possible in this, since an attacker who wants to outvote the system would only have to register a sufficient number of accounts in the network, which would be possible for him without significant costs (this is called a Sybil Attack, see e.g. [ 17]).

An unrestricted system like Bitcoin must therefore tie the weight of one vote when voting to a scarce resource in order to prevent such attacks. With Bitcoin and many other cryptocurrencies, this is done via the so-called Proof of Work (PoW), i.e. In other words, the weight of the voice is linked to proven calculation work and thus energy. The proof of arithmetic work consists in finding a random number, the so-called nonce, so that the hash value of the nonce - together with other data - takes on a certain form. In the case of Bitcoin, this is the requirement that the integer representation of the hash value is less than a certain upper limit. The choice of this upper limit defines a difficulty, the so-called difficulty, of this cryptographic puzzle. The difficulty is indirectly proportional to the probability that a randomly chosen nonce will lead to a hash value of the desired form. This method of verifying computing power has been known for a long time and was discussed in Hashcash for preventing spam [18].

Participation in the PoW consensus mechanism, i.e. the search for corresponding nonces, is therefore associated with costs, so that an economic incentive must be created for participation in mining: Whoever finds a nonce that, together with a bundle of transactions at a hash value, is required Form is allowed to enter a reward in the amount of a certain number of Bitcoins newly created for this purpose (Block Reward). The corresponding block can then be communicated to the other participants in the blockchain network and thus attached to the existing chain, which means that the corresponding transactions are carried out. It should be mentioned that due to the competition in Bitcoin mining, participation with CPUs has long since become unprofitable, as specialized hardware, so-called ASICs, have been developed that can calculate hashes many orders of magnitude faster and more energy-efficiently than CPUs and GPUs [3]. The difference with Bitcoin is so serious that even the 500 largest supercomputers in the world combined can probably only achieve a small part of the current Bitcoin hashrate, mostly on ASICs - and only with considerable financial losses.

In order to prevent undetected manipulations in the system, Bitcoin uses a data structure that makes it very easy to recognize and find subsequent changes, namely Merkle Trees. In order to reduce the effort involved in finding consensus, numerous transactions, metadata, nonce and a hash pointer to the previous block are combined in one block. The resulting append-only structure (chain) is given the property that changing just a single transaction either leads to the inconsistency of a single block (false Merkle root) or all hash pointers from the manipulated block would also have to be changed. Due to the requirement that the hash value of each block must have the form described above, finding such blocks and thus an alternative chain of hash pointers is very computationally expensive, so that the system is secure as long as a large part of the hashrate is provided by "honest" nodes ( for attack scenarios, see for example [19]). The data structure of blocks and hash pointers is generally characteristic of blockchain technology, which in turn is a special case of so-called distributed ledger technologies. Usually, however, blockchain technology is not only understood to mean this data structure, but also the existence of a consensus mechanism that enables both an agreement on the addition of new transactions and ensures that no subsequent changes can be made to the blockchain. For this, however, in addition to the difficulty used in PoW of finding archetypes of certain hash values, other methods can also be used. These are usually digital signatures that, depending on the consensus mechanism, are created by either fixed rules or (pseudo-) randomly determined network participants. We will briefly address these variants when discussing alternative consensus mechanisms.

Estimates for the energy consumption of PoW blockchains

As already described, the incentive for participating in mining in the Bitcoin blockchain and in general for PoW-based blockchains is the reward in the form of units of the associated ("native") cryptocurrency, which is required as an incentive system. Due to the strong price increases of crypto currencies with the peak at the end of 2017 and a market capitalization of briefly over 300 billion and since then always over 50 billion US dollars from Bitcoin alone, there is and was a great economic incentive for participation in mining. In order to maintain the functionality (and also the security) of the PoW blockchain network, the period of time in which a new block is usually found must be kept constant, i.e. That is, the difficulty of the hash puzzle must be adjusted according to the current hashrate. This leads to a correspondingly high power consumption of PoW-based cryptocurrencies.

Basically, the exact determination of the power consumption in an open and non-restricted PoW blockchain is very difficult, as it is usually not possible to determine the computing power used in mining or the corresponding hardware for each individual participant. However, a lower limit for the power consumption of Bitcoin and all other PoW-based blockchains can easily be determined from the indirectly observable average computing power, i.e. the hashrate, and the most energy-efficient mining hardware on the market [3, 20]. The expected value of the hashrate can be estimated from the publicly visible current difficulty of the hash puzzle and the number of solutions communicated in the form of new blocks. With Bitcoin, after the protocol has been constructed, a new solution to the hash puzzle is found on average every 10 minutes and the probability that a random hash value fulfills the requirements was around \ (1: 6 \ times 10 ^ {22} \) at the beginning of 2020. For Bitcoin, SHA-256 is used as the hashing algorithm; modern ASICs achieve hash rates in the order of magnitude of \ (10 ​​^ {14} \) hashes per second with an output of a few thousand watts. This gives a lower limit of around 60 TWh annually for Bitcoin's electricity requirement at the beginning of 2020 [11], which corresponds to the annual electricity consumption of around 15 million households.

An upper limit for the electricity consumption caused by mining can also be estimated, as long as one assumes that all participants in the consensus mechanism act rationally and strive for profit by participating in the mining. This may not be true for all participants, but the vast majority of the computing power for Bitcoin and other relevant PoW cryptocurrencies is provided by companies or groups (pools) specializing in mining [21], for which this assumption makes sense. The value of the economic incentive, i.e. the bitcoins newly generated by mining, must on average be at least as high as for the costs caused by the mining, e.g. for electrical energy and hardware, and thus in particular higher than the electricity costs. A lower limit for the costs of electrical energy in countries with significant participation in mining is usually set at 0.05 USD per kilowatt hour [22, 23] and this gives an upper limit for the start of 2020 at a bitcoin price of almost USD 10,000 Annual electricity consumption of around 120 TWh, which corresponds to around 20% of German electricity consumption [11].

For other, well-known PoW blockchains, such as Ethereum, Bitcoin Cash, Bitcoin SV and Litecoin (these are the largest PoW-based cryptocurrencies after Bitcoin by market capitalization), the same estimation formulas apply as for Bitcoin, only there are other hashing algorithms, specialized ones Mining hardware and parameters such as mean block times and block rewards. In total, the power consumption of the 4 cryptocurrencies mentioned is between 10 TWh and 40 TWh per year and is thus significantly lower than that of Bitcoin. It is also found that, due to generally similar parameters, there is a very high correlation between market capitalization and power consumption for different PoW cryptocurrencies. Since Bitcoin's market capitalization is higher than the cumulative of all other cryptocurrencies, it can be assumed that the cumulative power consumption of all PoW cryptocurrencies is not much more than twice that of Bitcoin, and a Best Guess is a factor of around 1 , 5 [4, 11].

An important observation for PoW cryptocurrencies is that their power consumption cannot be reduced in the long term by increasing the energy efficiency of hardware: On the one hand, this can be seen from the fact that the estimate of the upper limit only depends on the electricity prices and not on the computing power. The reason for this is that in the long term, all miners would switch to more energy-efficient hardware as long as mining is profitable with it. As already described, the overall computing power of the network increases accordingly until the balance between the income and expenditure side is approximately restored.

Because of the requirement to enable as many nodes as possible to participate in cryptocurrencies, as well as the redundant execution of all transactions, the technical requirements for participation, i.e. network bandwidth and storage space, must be kept as low as possible. Since the "slowest" permissible node determines the performance of the system, Bitcoin and other cryptocurrency systems can only process a few transactions per second - the storage space required for the complete Bitcoin blockchain currently requires almost 300 GB and grows by around 60 GB per year, one Multiplying transactions per unit of time would also multiply growth. Accordingly, simply dividing the power consumption by the number of transactions with PoW-based cryptocurrencies results in an enormous amount of energy per transaction: For Bitcoin, the power consumption for a single transaction is several hundred kWh and thus corresponds to the power consumption of an average German household from several weeks to Months, which leads to the frequent criticism of the sustainability of Bitcoin as described above. For other PoW-based cryptocurrencies, you get significantly lower values ​​per transaction, but this is still orders of magnitude more energy-intensive than, for example, a conventional booking in the banking system. However, it is essential to understand that the number of transactions processed has no effect on the power consumption of the overall network caused by the mining, since in theory the blocks could be enlarged at will [24]. Thus, the metric “energy per transaction” for PoW-based crypto currencies is to be considered ambivalent. Nevertheless, given the performance of Bitcoin and other current PoW blockchains, their power consumption can be described as disproportionate.

Outlook: Implications from the recent Bitcoin halving

In the following, a more detailed analysis of the power consumption of Bitcoin is carried out through an analysis of the recent Bitcoin Halving and implications for the long-term development of power consumption are derived from this. The comparison of the development of Bitcoin prices and hashrate over the past 12 months, shown in Fig. 1, suggests that the upper limit described above is actually a very good estimate of the actual power consumption: If Bitcoin prices are relatively stable until March 2020, it will increase observed hashrate continuously; evidently the start-up or expansion of mining activities, which is connected with the procurement of appropriate hardware, were seen as worthwhile. A fall in the price of Bitcoin at the beginning of March 2020 as part of a generally weak mood on the stock market as a result of the COVID-19 pandemic was accompanied by a slightly less pronounced, but nevertheless significant, drop in the hashrate. This could be explained by the fact that because of the reduction in the value of Bitcoin and thus the level of the mining incentive, miners with higher variable costs, for example due to outdated hardware or high electricity prices, were forced out of mining for a short time. After that, with the Bitcoin price, the hashrate rose again to the previous level. However, the Bitcoin halving on May 11, 2020, which is provided for in many PoW blockchains and, in the case of Bitcoin, happens about every 4 years, then resulted in a permanent halving of the block rewards and a corresponding reduction in the economic incentive for mining. Since Bitcoin prices remained largely constant, the hashrate fell significantly, similar to before.

Surprisingly, however, the hashrate rose significantly again just a week later, without the prices for Bitcoin rising to a similar extent. This could have the following causes:

  • A closer look at the expected profits from mining also includes transaction fees that the producer of a new block receives - especially after halving, these account for up to 20% of the reward on some days and the transaction fees are also partly clear after halving increased.

  • The difficulty is not adjusted in real time, but only about every 14 days - with the difficulty before the halving it may not be worthwhile to participate in the mining, but with the difficulty adjusted for the first time after the halving and thus significantly reduced (es So it takes a while for the system to regain balance).

  • In China, the rainy season begins in May and June in some regions, so that much cheaper electricity is available through hydropower and some mining pools already offer electricity for just under USD 0.03 / kWh, especially since the competition is clear in view of the falling mining revenues has tightened [26].

  • Modern, energy-efficient hardware is increasingly being acquired and used, which significantly reduces variable costs.

To investigate these relationships, we analyze the relationship between the economic incentive to participate in mining in the form of an expected mining revenue in USD per \ (10 ​​^ {\ mathsf {18}} \) calculated hashes and the actual participation in the form the hashrate, in the period around the halving with a more precise model. This also takes into account the actual current block time or difficulty as well as the transaction fees received from the generator of a new block in addition to the block rewards. The resulting courses are shown in Fig. 2. It shows that the correlation between expected profits and the hashrate is very high at approx. 0.57. This suggests that many miners are already deciding in real time or at short notice whether or not it is currently worthwhile for them to participate in mining, because they have already operated close to cost neutrality before halving. In addition to the irregularities around the halving, lower correlations can be recorded, especially in the last 2 weeks shown, this could be due to the described changed electricity tariffs in China's mining pools.

In order to analyze the influence of the electricity tariffs in more detail, we consider in Fig. 3 the influence of mining hardware and electricity prices on the relative margin, i.e. the ratio of mining profits (i.e. the difference between mining revenues and electricity costs) to electricity costs of mining in detail. Only electricity consumption is taken into account; further variable costs and investments (e.g. for hardware procurement) are ignored. As a result, the data shown correspond to an upper limit for the relative margin. The widely used Bitmain Antminer S9 (11.5 TH), for 2018 the MicroBT Whatsminer M10S and for 2020 the Bitmain Antminer S19 Pro (110 TH) were used as reference hardware for hardware that came onto the market in 2016 [27] . At the time of their market launch, these probably corresponded to the most energy-efficient hardware. As in Fig. 1 and Fig. 2, the vertical line shows the time of the Bitcoin halving. In fact, with electricity prices of USD 0.05 / kWh, old, less energy-efficient hardware is pushed out of the market for a short time by halving, whereas more modern, more energy-efficient hardware remains profitable and, at lower electricity prices, mining with older hardware also makes economic sense.

Based on the observations, it is obvious that the derived upper bound is now actually a good estimate for the actual power consumption of Bitcoin. Fig. 4 shows the power consumption of Bitcoin resulting from the different scenarios. The lower limit with 2020 hardware and the upper limit with 0.025 USD / kWh can be viewed as very reliable, as it can hardly be assumed that significant mining activities will take place with more efficient hardware than the most modern hardware available on the market or that electricity costs below 0.025 USD / kWh will take place . However, the two dashed barriers are more realistic on the basis of hardware and electricity prices that are actually typical in the network. The estimates by Digiconomist [28] and Cambridge [29] seem plausible in view of the safe upper and lower bounds and fit well into these realistic bounds, but may not adequately explain the collapse of the hashrate as a result of the price fall and the Bitcoin halving. In this respect, one could certainly expect that the actual hashrate before the halving was more oriented towards the upper limit of 0.05 USD / kWh and, after the halving, due to the increased competition from cheap electricity tariffs, mining hardware, which was initially pushed out of the market was used again and the actual power consumption has even risen above the upper limit of 0.05 USD / kWh.

One could expect, however, that this is only a temporary effect and that the electricity consumption will again be oriented towards the upper limit at 0.05 USD / kWh after the end of the rainy season. Even if the electricity consumption of Bitcoin should stay in the range of less than 0.025 USD / kWh due to very low electricity prices, it can be assumed that in this case it is electricity from renewable energies, since their marginal costs can also be 0, whereas the cost of electricity generation from fossil or nuclear fuels should hardly fall below USD 0.025 / kWh. In this respect, even exceeding the “safe” upper limit due to locally or temporarily lower electricity prices would probably not result in a higher CO2-Footprint for Bitcoin mean as with the upper bound with 0.025 USD / kWh.

Assuming constant economic conditions, i.e. the prices for Bitcoin and electricity as well as the transaction fees, this orientation of the actual electricity consumption to the upper limit also means that the electricity consumption of Bitcoin will decrease significantly in the long term due to the periodic halvings. Currently, the block rewards still make up around 80% of mining revenues; if the prices for electricity and Bitcoin and transaction fees remain the same, electricity consumption would be 40%, 60% and 70% in 4, 8 and 12 years respectively compared to the beginning of 2020 % decrease and in the long term settle at around 20% of today's value, i.e. 20 TWh / a and thus around 4% of today's German electricity consumption. Since economic framework conditions such as the prices for Bitcoin and transaction fees can practically not be reliably predicted, this does not represent a reliable forecast. However, one can conclude from this that in a few decades and with limited Bitcoin prices the value of the electrical energy consumed by Bitcoin will not be more will be significantly greater than the cumulative transaction fees. From around the year 2140 this will be the case anyway; Whether Bitcoin will be relevant for so long, however, can hardly be judged after only 11 years. Due to the competition with other crypto currencies as well as technical developments, for example with regard to performance (reducing the scarcity of transactions), one could assume that the transaction fees will not increase significantly compared to today's level. This argument can be applied to many cryptocurrencies, which, like Bitcoin, have a periodic halving of the block rewards.

Alternative consensus mechanisms

In general, the power consumption of blockchains consists of 2 components: the consensus mechanism, i.e. the process in which the nodes agree which transactions are to be executed in which order, and the redundant execution of the transactions themselves, i.e. the verification of signatures and the adjustment of "account balances" in the local database of each node (state transitions). With Bitcoin, the PoW consensus mechanism, as described, ensures the enormous power consumption; the cumulative effort for the redundant execution of the transactions is negligible even with the current size of the network of approx. 10,000 nodes [3]. Since the development of Bitcoin, however, alternative consensus mechanisms have been developed - partly because of the high power consumption of PoW. For non-restricted systems, the most successful alternative to date is the Proof-of-Stake (PoS) consensus mechanism. Here, the voting weight is not linked to the computing power provided, but to the scarce resource capital that is visible and therefore verifiable within the network, i.e. owned units of the cryptocurrency. Often capital has to be "frozen" for a certain period of time in order to participate in the vote and then serves as security to create an economic incentive for correct behavior of the nodes. With PoS there is no competition for more and more computing power. The energy requirement associated with the consensus mechanism is accordingly negligible compared to PoW and requires only a few operations on each node. Examples of current implementations of blockchains with a PoS consensus mechanism are Algorand, Cardano, EOS and Tezos. In addition, a comprehensive further development, Ethereum 2.0 (Serenity), which is also based on PoS and should replace or integrate the current Ethereum blockchain in the long term, will go into operation for Ethereum, the second largest cryptocurrency in terms of market capitalization. . In addition to purely PoS-based blockchains, there are also combinations of PoS and choice-based protocols, such as the consensus mechanism Tendermint used for the Cosmos network. Such protocols also do not contain a particularly computational or energy-intensive process.

There are often discussions among supporters of PoW and PoS as to whether PoS is just as secure as PoW - there are good arguments for both sides. For example, centralization effects are to be expected at PoW in the long term due to economies of scale in mining (procurement of hardware and electrical energy) and location-dependent economic framework conditions, which confirms the currently strong centralization of Bitcoin mining. This in turn can lead to security problems. With PoS, on the other hand, participation in the consensus mechanism corresponds to a return on the capital employed, so that the ratio of the capital of all participants in the consensus mechanism and thus also the weight of their votes remain constant. On the other hand, with PoW you can also participate in the consensus without first having to get resources from the network itself, as with PoS, which could make redecentralization significantly more difficult in an already highly centralized system. In addition, many PoS systems currently only allow participation in the consensus mechanism from a certain minimum deposit and for PoW blockchains there is also a simpler decision rule as to which blockchain is the "valid" one in the event of a conflict (forks). The success of PoS blockchains in recent years and research in this area [31] indicate that PoS can guarantee a level of security that is comparable to that of PoW.

Access-restricted blockchains used in the context of consortia in the private or public sector all use choice-based consensus mechanisms, some of which can be viewed as crash-fault-tolerant simplifications (e.g. RAFT) or byzantine-fault-tolerant (e.g. PBFT or RBFT) successors to Paxos . Their consensus-based energy consumption is therefore negligibly small, as is the case with PoS.

Redundant operations

Regardless of the type of consensus mechanisms, all blockchains are characterized by redundant data storage and operations. Accordingly, the associated cumulative computing effort and thus power consumption - if one disregards hardware-related differences - is proportional to the number of participants in the blockchain network. For small blockchain networks, as they are typically used in a limited-access context in consortia, the redundancy results in a multiple power consumption compared to a central system. However, this does not have to mean that the use of a blockchain has to be negative from a sustainability perspective. The following rough estimate should illustrate this: A small, private blockchain network, such as Hyperledger Fabric or Quorum, with 10 nodes and a mediocre hardware configuration (2 CPUs, 8 GB RAM) can easily process 1000 simple transactions per second. This means an energy consumption of a maximum of 1 year per transaction. On the other hand, it can be calculated from the information given in VISA's 2017 sustainability report that the energy consumption of the entire company (including heating buildings, etc.) is calculated down to approx. 6000 J per transaction, of which approx. 3000 J per transaction are incurred for data centers [32]. A simple client-server system with a simple key-value store, such as LevelDB, can process several thousand simple transactions per second with the hardware equipment mentioned above, which leads to a power consumption of around 0.02 J per transaction. Accordingly, the power consumption of a blockchain is usually significantly higher than that of a corresponding central solution (here by a factor of 50) due to the redundancy (and partly also due to the consensus as well as the more extensive use of cryptographic methods), but it may Nevertheless, only a very small part of the power consumption of the entire IT solution or the entire process is used, even if clients and backups are also included. Especially in scenarios in which processes can be further digitized with the help of energy-efficient variants of blockchain technology, it is not absurd that blockchain-based solutions can ultimately save energy.

In the most well-known cryptocurrencies, such as Bitcoin and Ethereum, the corresponding blockchain networks already consist of many thousands of nodes and their number is likely to increase sharply in the future with large-scale adoption. Accordingly, the power consumption of these networks can be viewed as problematic because of the resulting high degree of redundancy. However, research has promising solution concepts in store for this challenge as well: In principle, a reduced degree of redundancy, i.e. the recalculation of transactions only on subsets of all nodes, also reduces the power consumption per transaction. This is the case with so-called second-layer concepts such as lightening or raiding, but typically associated with trade-offs in terms of security, as this is based on the high degree of redundancy. Similarly, in the case of the Ethereum 2.0 blockchain already mentioned, the network will be divided into a total of 64 so-called shards, which are periodically referenced by a main chain, the so-called beacon chain, and thus inherit the security of the overall system with each such referencing. It will be some time before Ethereum 2.0 uses these features to their full extent and processes are distributed accordingly across the various shards, so it is currently difficult to quantify how much the degree of redundancy will ultimately decrease and what impact this will have on security and functionality becomes.

ZK rollups

Particularly promising are current advances in connection with "Proofs of Computational Integrity" that have come into focus in recent years - especially through blockchain technology - which are perhaps better described in the blockchain environment under the catchphrase "Zero Knowledge Proofs" (ZKP) knows. This makes it possible to show (probabilistically) with a usually very short and easily verifiable proof that certain calculations have been carried out correctly without having to specify all the details of the calculation or having to repeat the complete calculation. Initially, ZKP were used by some cryptocurrencies such as Zcash in order to restore the confidentiality of transactions, which practically does not exist with cryptocurrencies such as Bitcoin [33].

An essential property of typical ZKP is that the size of the evidence and the computational complexity of the verification of the evidence usually scale sublinearly (e.g. constant or polylogarithmic with SNARKS or STARKS [34, 35]) with the size of the calculation to be verified . This makes it possible for a single party, for example a crypto exchange, to bundle a large number (several thousand or tens of thousands of transactions) in so-called ZK rollups and only provide brief evidence that all steps have been carried out correctly (checking of signatures, correct updating of Account balances ...), sent as a transaction to the blockchain, where the evidence is checked by the other nodes with little effort. In detail, the architectures can differ significantly here, both in terms of the ZKP technology used and the other data stored on the blockchain and continuously updated. (There is a tradeoff here: Less data on the blockchain means a higher dependency on the party who creates the evidence and updates and stores the account balances of all accounts, but also higher scalability, since the blockchain then does not represent a bottleneck.)

In contrast to previously existing second-layer solutions, ZK rollups with complete on-chain data storage can guarantee the same security guarantees as the blockchain itself, since the evidence is still checked by all participants in the network and manipulations are therefore as secure as conventional ones Transactions can be excluded [36]. In addition, significant improvements can be achieved compared to the conventional processing of transactions, since the majority of the storage and computing capacity is based on digital signatures, the verification of which the operator of the ZK rollup compresses into a brief proof. In existing prototypes, transaction rates of several hundred to several thousand transactions per second in Ethereum (conventionally approx. 10 transactions per second) are already achieved with the help of ZK rollups [37], [38]. We want to use an example to estimate what implications this can have on the redundancy-related share of the power consumption of a large blockchain network.

On the basis of the ZK roll-up from Loopring, which is already in use on the Ethereum blockchain, the potential for savings in the ideal case, i.e. with maximum utilization of the ZK roll-up and only transactions within the ZK roll-up, can be estimated well: For Loopring 3 the so-called gas costs, which are a measure of the storage and computing effort of a transaction in Ethereum and determine the pricing of the execution of transactions, are put at 365 at a maximum load of 2100 transactions per second [37]. For comparison, a simple transaction in Ethereum requires at least 21,000 gas, often significantly more. Accordingly, this already means a reduction in redundant operations and thus their power requirements per transaction by a factor of approx. 100 (there is no exact proportionality between gas costs and computing effort, but this is approximately a reasonable assumption). On the other hand, the cost of the computationally expensive creation of evidence, which is only necessary for the operator of the roll-up, is estimated at USD 0.000042 per transaction [37]. With 2100 transactions per second, this corresponds to an amount of approx. 5 USD per minute and thus, for example, the operation of AWS instances with approx. 96 vCPUs. For this one can estimate a power consumption of a few hundred watts [39, 40], which does not result in more than 0.5 J per transaction. This also corresponds to the order of magnitude from an alternative estimate, which indicates, at least for ZKP-friendly hash functions, an effort for the creation of evidence that is increased by a factor of 100 compared to a simple calculation. On the other hand, our own measurements with ethereum-based blockchains without a PoW consensus mechanism result in a power consumption of approx. 0.01 J per transaction (tx) and node from CPU usage. In a network of around 10,000 nodes, which roughly corresponds to the size of the Bitcoin and Ethereum networks today, this results in a power consumption of around 100 J / tx without ZK roll-up. With ZK rollup, on the other hand, you get an amount of (100/100 + 0 {,} 5) J / tx and thus an energy saving of 98.5 for the redundant operations, i.e. primarily the verification of the proof %. For even larger networks, the energy savings in our example would increase accordingly to up to 99%, whereas they would decrease for smaller networks and even lead to an increase in power consumption for networks with very few nodes.

However, in a holistic view, the no-load current of nodes must also be taken into account. Depending on whether a separate PC is used to participate in the blockchain or whether it remains switched on or the PC is running anyway or is located in a cloud with a tendency towards low idle current, as well as depending on the average load on the blockchain network, the idle current can be for large networks in non-PoW blockchains, the amounts of energy used in connection with transactions can be significantly higher or negligible. Further improvements in energy proportionality could mean that no-load current will be less important in the future [41]; large data centers are usually more advanced than small ones [42]. In addition, a scenario in which all transactions are processed within an access control roll-up is unrealistic. Nevertheless, it is conceivable that a large part of financial transactions can be processed within such ZK rollups. ZK rollups were primarily developed to solve scalability and performance problems of blockchains and, as just described, have the pleasant side effect that they can also contribute to considerable improvements in power consumption. However, these are only noticeable if no PoW-based consensus mechanism already causes such a high power consumption that improvements in the redundant operations are not noticeable at all, and the idle power is absolutely negligible or in relation to the power consumption from CPU usage caused by transactions is.

Conclusion

The PoW blockchains on which numerous cryptocurrencies are currently based and in particular Bitcoin - taking into account their current technical performance - consume enormous amounts of energy. The total electricity consumption of all these PoW cryptocurrencies is still largely caused by Bitcoin and amounts to between 20-50% of German electricity consumption, with a best guess for Bitcoin at around 100 TWh / a or 20% of German electricity consumption. The driving force behind the electricity consumption is the price of Bitcoin and not the number of transactions and if the economic framework conditions remain the same, the corresponding electricity consumption would also decrease significantly in the long term due to the periodic halving of the block rewards that occurs with many PoW-based cryptocurrencies.

In addition, there are now established blockchains with alternative consensus mechanisms, especially PoS for open, non-restricted cryptocurrencies and the choice-based consensus mechanisms of private, restricted-access blockchains. The latter usually include solutions that are used, for example, in companies or the public sector as a cross-organizational, neutral platform. Due to the elimination of the PoW, their power consumption is in each case orders of magnitude lower than that of Bitcoin and other PoW-based cryptocurrencies. Mainly due to the redundant calculations that are characteristic of blockchain technology, however, their power consumption per transaction is roughly proportional to the number of participating nodes and thus still many times higher than that of central systems. For large cryptocurrency networks in particular, this can still mean high power consumption even for non-PoW blockchains. Technological developments and modifications, with which the effort for redundant calculations and data storage can be reduced, and in particular zero-knowledge proofs in ZK rollups, can be expected that the power consumption of large networks can be significantly reduced again in the future. In a holistic view, however, the no-load current must always be taken into account.

Fig. 5 summarizes the findings of this article by estimating the typical power consumption of the different blockchain technologies and improvements described in the previous sections. These are primarily for illustrative purposes and do not take into account the no-load current. The numerical values ​​given should therefore not be regarded as fixed and reliable, but only as an indication of the order of magnitude. In particular, the error estimates are in most cases not generally reliable, as they correspond to empirical values ​​from tests with different systems. However, the orders of magnitude are reasonable estimates based on the assumptions made in the respective chapters to the best of the authors' knowledge. Fig. 5 illustrates that the power consumption of the blockchains already used in the application today is reduced by several orders of magnitude compared to the PoW blockchains of the "first generation" or could be reduced with the technology available today.

Even if the use of blockchain technology may not be the most energy-efficient solution from a purely technical point of view, it ultimately depends on what energy savings can be achieved in return through the use of the technology. The potential in the cross-branch and cross-branch use of IT is particularly great [43]. It is precisely here that many of the blockchain applications that go beyond cryptocurrencies and that are currently being developed or tested in business and the public sector come into play. Because for economic or political reasons, it is often not possible to implement a central, digital platform in these scenarios. When using the energy-efficient blockchain solutions described, it can be assumed that the associated automation of processes can save resources not only from a financial point of view, but also from an energetic point of view.

Here it is the task of business informatics to identify and quantify the existing potential for energy savings and climate protection and - depending on the scenario with or without blockchain technology - to leverage it with the most suitable technology. The prerequisite for this is that the consumption of resources (e.g. by a CO2Tax) is priced in such a way that no ecologically harmful distortions occur and therefore the economic incentive mechanisms promote the development of such solutions. In addition, one should always be aware that a solution that initially appears expensive due to the high complexity and the still young technology, which reduces the overall consumption of resources through the use of modern solutions such as blockchain, is also a safeguard against future price increases or fluctuations can [44].

In conclusion, we therefore want to encourage further research both in the area of ​​technical improvements of blockchain technology, for example with regard to performance and energy efficiency, as well as areas of application with particularly high potential for energy savings.

literature

  1. 1.

    Thiel T (2020) Viruses in the Gold Rush.