Just how energy efficient is your blockchain?
Sustainability is on everyone’s minds these days. Take electric vehicles – nearly every auto manufacturer is producing one; many governments subsidize them; and the infrastructure for them is growing quickly. For the first time, cars are steadily becoming less polluting.
How do different blockchains compare? With headlines focusing on the sustainability problem of Bitcoin, you may be wondering what other options are out there, and whether they are really any better. The main contributing factor to energy consumption is the consensus process. But this can look very different depending on which blockchain you’re measuring. We decided to do some quick math to see if we could come up with a like-for-like ranking.
Bitcoin’s consensus problem
Both Bitcoin and large heavy cars have a reputation for being terrible for the environment. Every journey and transaction requires large amounts of energy. For that gas guzzler speeding down the highway, the engine is the problem; it uses fossil fuels which are non-renewable and pollute the environment.
For blockchain, the problematic feature is the consensus mechanism. Consensus is how a blockchain validates a transaction. While different blockchains use different consensus mechanisms, Bitcoin’s popularity and reach makes its mechanism – called Proof of Work (PoW) – the Hummer of blockchains.
In PoW, users compete to mine blocks of transactions. Mining is a competitive part of the validation process used to ensure that the input of a transaction is the same as its output. The process of finding an integer that results in the hash value of the block being under the current target difficulty requires many attempts, by many competing miners, which translates to a lot of computing power. Additionally, the market capitalization of a cryptocurrency strongly correlates with its energy demand, making any proof-of-work based cryptocurrency anti-efficient from an energy consumption perspective. This is particularly evident in popular cryptocurrencies like Bitcoin.
But wait! Every blockchain works a little differently
But not every blockchain is a Hummer. PoW is only one of several common consensus mechanisms. Others are Proof of Stake (PoS) and the low-energy alternatives of private blockchains.
In comparison to the energy-intensive mining free-for-all of PoW, PoS gives those with larger cryptocurrency holdings more power in transaction validation. Depending on the design of the PoS protocol, a validator may stake cryptocurrency in order to qualify as a validator. Should they abuse this position of power by acting dishonestly, the staked amounts may be “slashed”. This motivates good behavior, which obviates the need for computing puzzles in consensus, meaning that less energy is consumed than in PoW models.
Achieving finality in Corda is even more efficient
A broad approximation of the energy needed to achieve transaction finality can be made by taking into account the energy demand of servers responsible for transaction finalisation. In PoW dedicated mining hardware with a large energy footprint is required and, depending on the popularity of the blockchain, a large number of miners may participate. In PoS, commodity hardware is sufficient. Still, the number of participating validator machines can be high. In the Corda protocol, where notary nodes are responsible for transaction finalisation, networks can operate with a small number of validator nodes. In fact, most deployments, including the public Corda network, utilise just a single logical notary.
Energy consumption of different blockchains
| Blockchain | Transaction Finalisation | Energy consumed per transaction (Joules/Tx) |
|---|---|---|
| Bitcoin | Proof of work | 125,000,000 |
| Ethereum | Proof of work | 17,222,222 |
| Polkadot | Proof of stake | 1663 |
| Corda | Corda Notary Service | 24.6 |
Note: data in this chart is based on publicly available data and may not reflect exact volumes.
Sources: Data for Bitcoin, Ethereum and Polkadot sourced from Powell, Hendon Mangle and Wimmer (2021). Data for Corda calculated from the Corda Network using the formula (205*6)/50.
For our purposes here – as a back of the envelope effort to compare magnitudes only – a single number “snapshot in time” approach was used. But as any holder of cryptocurrency knows, the number of validators and throughput changes frequently for public blockchains. Conceivably, the number of validators may even be correlated with the number of transactions on a blockchain. We believe that the table functions as a reasonable estimate of comparable magnitudes.
We have a couple of caveats to this table. In an ideal world, we would have high quality public data for each variable. In reality, the quality is uneven. We made our best efforts to use publicly available published data.
Second, we make some effort to overestimate Corda’s energy consumption. The estimates are driven down by either very high TPS or low numbers of servers. We use higher values when there is a range and the least efficient architecture where there is a selection. For example, the server energy consumption is at the high end of the spectrum, and the TPS we use takes the current average of 50 rather than the demonstrated potential for 6,300 TPS under a specialized configuration. This is intended to balance any perceived bias on the part of the authors.
But is this really comparing like-for-like?
We made our best effort to compare the same part of the consensus actions for all of the blockchains. Most existing work on energy usage measures public blockchains and the mining process. To extend this to Corda, which does not have mining, we needed to make some assumptions.
The biggest comparison hurdle was the definition of a transaction. In the case of public blockchains, this is clearly the achievement of global consensus about a particular set of UTXOs. This is a highly secure process in bitcoin. To extend this to Corda, which uses a notary service, we assume a simple payment transaction between two parties that processes a single state. Where a bitcoin transaction is always a payment, a Corda transaction can be anything from a smart contract execution to an asset transfer.
The calculation
There are differences in the calculations used for PoW, PoS and the Corda Notary service. The reason is that the protocols, number of validator nodes, recommended hardware, and need for throughput vary.
A very rough model we used to calculate energy use was this one:
Energy per transaction = (power per server x number of servers) / transactions per second
We adopted this admittedly highly simplified model from a peer-reviewed paper on the industry’s efforts to make blockchain greener. It assumes that a server consumes a certain amount of energy, that there are a known number of servers needed to operate a blockchain network, and that blockchain networks have a known volume of throughput.
For the public blockchains, their energy consumption figures were converted to watts per second, and then divisible by transactions per second (see figures for Bitcoin, Polkadot, Ethereum). This yielded numbers that are comparable to others published about these blockchains.
Power per server – The savvy reader will know that server consumption is a tricky variable. A recent paper on blockchain consensus mechanisms estimates that it can take a range of value from 5.5 to 328 watts depending on hardware type.
For Bitcoin and Ethereum, the Cambridge Bitcoin Electricity Consumption Index calculates a detailed number that takes into account the type of mining equipment. For the other blockchains, we need to make assumptions about the type of equipment used. Polkadot numbers assume a Dell R730 rack-mounted server which uses 168 watt hours as their figure. For Corda, we take a number at the higher end of the spectrum at 205.
Number of servers – Again, for public blockchains this can be estimated directly as the hash rate is known. Polkadot also publishes these numbers, which for them was 732 according to their public site at the time of access. For Corda, we estimated the number of servers by assuming every notary processes one state per transaction (this is a lower bound of efficiency as most nodes are probably processing multiple states) and uses all 3 of the notaries on the Corda Network plus each of their associated database nodes (for a total of 6 nodes per transaction).
An important caveat is that all blockchains use multiple phases for a transaction to validate and finalize. A recent paper breaks it down into five steps for public blockchains. Since we employ existing calculations, we follow the practice of only including transaction finality processes in our calculations and do not include the validation phase (signing and sending) for Corda or any of the other blockchains in Table 1..
Transactions per second – there are considerable differences in transactions per second (TPS) between different blockchains. TPS is defined by the way transaction finality occurs in each blockchain. In table 1 above PoW generally has lower TPS than PoS and finally Corda. We use average numbers for the network at the time of writing. But In all cases, protocol parameters can be adjusted to achieve higher TPS. For Corda, we estimate 50 TPS average per Corda node using data from existing traffic on the Corda Network.
Sustainability and green blockchain use cases are on the rise
Increasingly, the blockchain community is seeking ways to improve global sustainability. The technology facilitates use cases that prioritise climate resilience, including supply chain and energy provenance and green finance. And it consolidates processes of legacy systems, making them much more energy efficient. By storing information on a blockchain, it becomes immutable and viable, which ensures that parties are held accountable for their green pledges.
Private blockchains are the most energy efficient choice that you can make today. This has led the way in promoting change throughout the community, as, just as you can’t really justify driving a Hummer anymore, it is no longer possible to argue that high energy consumption is the only way to run a blockchain.
Relevant links
