Dramatically Reducing the Carbon Footprint of Cloud Storage and Our Pledge to Become POW Free
When we started Storj, some of our primary motivations were environmental. We wanted to create a system that would:
Be environmentally friendly compared to both cloud storage and local storage.
Provide an environmentally friendly way for people to participate in the crypto economy.
In recent months, there has been increasing attention on the environmental impact of cryptocurrency systems. In particular, Proof of Work (PoW) systems can end up consuming huge amounts of power and equipment. PoW systems fundamentally rely on network participants and heavy machinery to solve cryptographic problems of increasing difficulty and, as a result, set off an arms race for miners to assemble ever more specialized equipment in massive data centers consuming ever-increasing amounts of cheap power.
To be fair, PoW blockchains also play an enormously valuable and beneficial role across many sectors, and helped spur the creation of multiple industries and opportunities for empowering people around the world. There is also considerable debate about how much electricity is consumed by large PoW systems, how much of that comes from renewable sources, what the true carbon footprint of PoW is, and whether the societal benefits of certain PoW systems outweigh the environmental costs, etc. However, even with the more conservative estimates, the potential environmental impact of PoW is too big to be ignored.
Surprisingly, even the higher estimates of the environmental impact of PoW pale in comparison to the environmental cost of cloud storage. Data centers are expected to account for more than 3% of global carbon output by 2025. By 2040, storing digital data is estimated to create 14 percent of the world’s emissions, around the same proportion as all carbon the US emits today.
When building Storj, we made multiple design considerations to enable the delivery of secure, durable, and performant storage in a manner that is significantly more carbon-friendly than traditional local and cloud storage. By design, our system is built to use existing unused capacity on already deployed hard drives and storage systems globally in a power- and resource-friendly way. Indeed, a recent survey says that 69% of Storj DCS capacity is deployed on hardware originally intended for other purposes. (Source: 2021 Storage Node Operator Survey)
We also made design decisions that minimized any wasteful use of PoW-based blockchain.
These design decisions were not purely altruistic. Making our system environmentally friendly also made the system more economically attractive for customers, storage node operators, and Storj itself.
While we are proud of our environmental record and what we have achieved to date, there is more work to be done.
Though we have minimized our use of PoW-blockchain over the years, there are still some aspects of Storj that use PoW-blockchain: most importantly, the STORJ token itself is an ERC-20 token that sits on top of the Ethereum blockchain. Ethereum is currently a PoW system, but is slated to move to the far more environmentally friendly Proof of Stake (PoS) approach in 2022.
This blog post covers four topics:
- The work that we have done to date to create a green platform
- Remaining technical work to further minimize our carbon footprint
- Our pledge to be PoW-free by the end of 2022
- Implementing a carbon offset program to mitigate both past and future carbon impact
1) Work to Date
Carbon Efficiency of Being a Storage Node Operator (SNO) relative to Mining
As noted above, environmental concerns were a primary motivator from Storj’s inception. While mining Bitcoin in his dorm room at Morehouse College, Storj founder Shawn Wilkinson hit upon a way to enable a system where under-utilized computing resources were used to provide true utility, rather than consuming scarce resources to solve increasingly difficult mathematical problems.
In the years since Storj's inception, our approach has proven out against many alternatives. Our platform stores real data that is useful to our customers and does not fill drives with cryptographic seed data. Our platform does not wear out hard drives prematurely, and data access is only the minimum required by use and health checks. Our platform can be used on affordable and older hardware and can even be run on Raspberry Pis, which is in stark contrast to other providers that require the latest and greatest hardware, causing supply chain issues and other downstream effects.
Carbon Efficiency Relative to Local Storage
As we designed and built the various versions of our product, it became clear that enabling the monetization and use of under-utilized storage resources would yield both tremendous environmental and economic benefits. Many disk drives are severely underutilized. Furthermore, a significant contributor to the lifetime carbon footprint of a drive appears to come from manufacturing and shipping the drive itself.
Thus, simply enabling more data to be stored on existing drives represents a major environmental advantage. Finally, since the vast majority of data remains static, it also requires little to no incremental power to keep a drive at 80% capacity vs. 25% capacity. Thus, if we can enable a terabyte of storage to be stored on an existing drive running with spare capacity rather than using a new drive, we can dramatically reduce the carbon footprint of that storage action.
Carbon Efficiency Relative to Traditional, Enterprise Cloud Storage
These benefits become even more stark when considering the use of decentralized storage vs centralized cloud storage. While centralized cloud storage providers do tend to make more efficient use of drives than local storage, other data center necessities such as cooling are much more costly and as a result, the carbon footprint of a data center is immense. Providing the physical plant, equipment, facilities, cooling, power, and backup power is incredibly capital- and carbon-intensive, according to one article:
“Recent predictions state that the energy consumption of data centres is set to account for 3.2 percent of the total worldwide carbon emissions by 2025 and they could consume no less than a fifth of global electricity. By 2040, storing digital data is set to create 14 percent of the world’s emissions, around the same proportion as the US does today.”
By contrast, Storj has been able to create the storage capacity of multiple data centers without knocking down any trees, hammering a single nail, digging a single field, pouring a single bag of concrete, or buying diesel for a single generator.
Carbon Efficiency in Achieving Durability/Resilience/Performance
The carbon efficiency story becomes even greater when one considers the steps necessary to achieve redundancy and geographic dispersion. Most traditional local and cloud systems (and even some other decentralized systems) attempt to achieve enterprise levels of durability through replication—i.e. by storing multiple copies of the same file. Given the mean time between failure of most drives, it would take 9-10 copies of a single file to achieve the 11 9’s of durability that Storj delivers (see Storj Whitepaper, section 3.4). That expansion factor of 9 means that a given terabyte of enterprise data incurs 9 times the cost and creates 9 times the carbon footprint of storing a single terabyte stored locally.
Furthermore, since many events (e.g. fires, floods, storms, power outages) impact all the drives in a data center, to achieve true resilience, high availability use cases require those extra copies to be stored in multiple different data centers in different geographic regions.
Storj has taken a different approach. First, we utilize erasure coding to achieve resilience. After segments of a file are encrypted, they are erasure coded. Typically, we utilize a scheme in which the encrypted file is split into 80 pieces, of which any 29 can be used to reconstitute the file. Each of those 80 pieces goes to a different drive on the network (run by different people, in different geographic regions, on different power supplies and ISPs), etc. The Storj scheme has an expansion factor of (80/29)=2.7, less than 1/3 of the cost and carbon footprint of the replication based system described above. Furthermore, by design, we also achieve geographic dispersion without building or powering additional data centers.
Design decision to minimize blockchain and PoW blockchain reliance
As should be clear, because we wanted to make it easy and economically viable to run nodes all around the world, and because we wanted to make it profitable and possible to utilize existing, underutilized capacity, we intentionally did not design a system which imposed significant overhead, cryptographic or otherwise. Even non-PoW blockchains (e.g. proof of replication, proof of space time) can create significant economic and environmental overhead. We chose to intentionally use an ERC-20 token for crypto-economics, rather than creating our own blockchain or our own community of miners.
To earn tokens, storage node operators need only: a) run a disk drive reliably, with sufficient levels of uptime, and b) not delete, corrupt, or lose the pieces of data that their node is supposed to store. Uptime is assessed by a continual set of lightweight “pings.” Data maintenance is determined by automated frequent random cryptographic audits which ask the node to return data derived from stripes of the pieces of the data that they are supposed to be storing. (see sections 4.13 and 5.7 of the white paper). By design, this is computationally easy for nodes that are storing the data that they are supposed to, but exceptionally challenging if data has been deleted or corrupted. As a node’s reputation improves, more data is stored on it.
The system thus rewards people who are generally good operators, not people who corner the market on cheap power or specialized ASICs. There is no need to form vast pools. As noted above, these pools have non-environmental benefits. Making it easier to become a node has the benefit of creating better economics, and enabling a diverse, low-concentration, highly-geographic-distributed population of nodes—which in turn improves durability and speed. The automatic audit system also has the benefit of detecting and repairing bitrot.
Storage Nodes on average have 1.48 TB free, and our used space is similarly distributed. And, regarding distribution, 50% of nodes are operated by people with three or fewer nodes total. To go further, about 70% of nodes are operated by people with fifteen or fewer nodes, which accounts for more than 75% of the data in the network. And, the Storj node network is well dispersed across more than 100 countries.
2) Remaining Work
While running a token on top of a PoW blockchain is less environmentally costly than running a separate blockchain, it is not costless. It is estimated that a single transaction on Ethereum consumes about 39 kg of carbon. While this number is also subject to considerable debate, it still points out the importance of the Ethereum move to PoS and the importance of reducing Ethereum transactions in the meantime.
We have adopted an off-chain layer 2 transaction rollup system, zkSync, for most of our low dollar value Storage Node Operator (SNO) transactions. In a rollup, all funds are held by a smart contract on Ethereum, while computation and storage are performed off-chain. The main idea is that instead of verifying each transaction separately, transactions are "rolled up" to a single item (the rollup block), which is then verified and approved simultaneously. For operators who are not ready to adopt zkSync, when the amount of STORJ owed to a SNO exceeds a certain dollar value we will push an actual blockchain transaction, but the combination of these two approaches has reduced our transaction volume by 86%. Supporting rollups via zkSync has also helped both Storj and our SNOs save on Ethereum gas fees. In addition, we are working to add more low cost Ethereum scaling options in the near future. For more on this, see our blog post on zkSync.
Similarly, there is more work to do to reduce the carbon footprint of storage operations.
We are evaluating our choice of Reed Solomon Erasure scheme. It may be that, for certain use cases, we can use ratios that result in lower expansion factors. Over time, that will reduce our carbon footprint further.
We also have a list of tools that are being built to help both users and Storage Node Operators use resources more efficiently. Again, this is an example of doing things that make both economic and environmental sense.
3) Our Pledge to Become POW Free in 2022
Without minimizing the enormous value of decentralized systems generally, and without minimizing the beneficial role that PoW systems like Bitcoin and the current version of Ethereum have played, we think the only responsible thing for us as a company is to minimize our own use of PoW where there are more environmentally friendly alternatives. To help play our role, we have pledged to make every effort to become PoW free by the end of 2022. Of course, the most important element of this will be the move to Ethereum’s new ‘Consensus Layer’ (formerly Ethereum 2.0). But, we are also investigating alternatives to our (relatively minor) use of PoW in preventing Sybil attacks in the node identity establishment process. (See whitepaper section 4.4).
4) Carbon Offset
While we can minimize our use of PoW systems going forward, we recognize that our historical use of STORJ on the existing PoW Ethereum network has had an environmental impact.
The wallets under our control had processed a total of 222,545 transactions as of January 6, 2022. Using an estimate of 14.1 kg amortized CO2e (carbon dioxide equivalent) emitted per transaction, that provides a carbon footprint estimate of 3,146 metric tonnes from all Storj operations. Since STORJ itself is traded, however, the carbon footprint extends to all transactions involving STORJ on the Ethereum Blockchain.
Why are we using 14.1kg CO2e? Offsetra did a fantastic analysis in 2020, (updated in early 2021), of carbon emissions due to Ethereum. They arrived at 0.18 grams of CO2e for every gas unit used by Ethereum, which roughly translates into 7.0kg CO2e per transaction for the types of transactions we’ve done.
Unfortunately, early 2021 is ages ago when it comes to these numbers. Some have argued that the increase in volumes and prices across Ethereum have almost certainly increased its carbon intensity. Again, without wading into the debate, we felt it was appropriate to multiply Offsetra’s 2021 numbers by 2. This may be an overestimate, as we are assuming our 2022 carbon emission estimates over the entire transaction history.
Based on recent data, at the time of this writing, it appears that there have been approximately 1.7 million transactions involving the STORJ token contract since its inception. You can see more on https://etherscan.io/address/0xb64ef51c888972c908cfacf59b47c1afbc0ab8ac
The average STORJ transaction, in general, seems to take slightly less gas than our average transaction, reducing the per-transaction estimate to about 13.7 kg CO2e, but overall, all STORJ transactions represent 23,495 metric tonnes of carbon to date.
To offset this, we are purchasing carbon offsets representing these 23,495 metric tons of carbon. We will update this post with a link to the STORJ transaction on Etherscan representing this purchase when it completes.
For the past 16 quarters, we have produced an accounting of token uses and flows. We think that it is only appropriate to include a carbon accounting to this report going forward. In addition, we pledge to make a quarterly offset as well on an ongoing basis for all STORJ transactions made by anyone. We want all STORJ transactions to remain green, whether or not we are making them - making us the most net-zero token and storage infrastructure available today - and beyond!
While we are proud of the work that we have done to date, and are excited about the potential for Storj to dramatically reduce the carbon footprint of cloud data storage, much work remains to be done. Fortunately, being green also makes economic sense. Our efforts to become more environmentally friendly also enable us to deliver better economics for customers, for storage node operators, and for Storj itself.