status

Cardano infrastructure under scrutiny

Published 9.6.2023

The Messari team released a report comparing several modern blockchain networks from the validator decentralization perspective. The report includes Cardano, Avalanche, Near, Solana, and Aptos. Unfortunately, the comparison with Ethereum is missing. The report is definitely worth a look, although from our perspective we think some of the information is not accurate. We hope to provide valuable context to the information in the report.

TLDR

Most validators are run in hosting services. If all validators (pools) running in AWS go offline, Cardano will be able to produce blocks. Many networks with instant transaction finality will not. The Cardano ecosystem has a large share of self-hosted validators.

Decentralization of stakes and infrastructure

The report focuses on the evaluation of validator decentralization. It examines geographic and infrastructure distribution in Proof-of-Stake networks. The team was particularly interested in the distribution of validators and stakes by geographic jurisdiction, and hosting infrastructure.

Every existing blockchain network is dependent on the Internet as it allows nodes to communicate with each other and periodically reach a network consensus (add a new block with transactions).

Every node operator has to make infrastructure decisions. This decision is especially crucial for those running block-producer nodes. In the Cardano network, these are called SPOs, or pools. The report talks about validators. The validator must ideally run without interruption, so the operator must ensure a 100% reliable connection to the internet and electricity. Some sort of failover solution is required. For example, once the internet connection goes down, the role of the node must be taken over by a redundant node at another location that has not lost connectivity.

Ordinary users running full nodes do not have to deal with this. Operators of block-producer nodes are responsible for producing blocks and validating blocks produced by other nodes. They are usually entitled to get rewards for this activity. In some cases, they may even be punished for poor service (slashing).

Validator operators usually use cloud services, as providers guarantee almost 100% reliability and protection against network attacks. Running a block-producer node from your own home can be costly and time-consuming. The monthly cost of running a block production node can be up to USD 1,000 per month if it is to provide similar quality to that offered by cloud services. These costs include regular hardware replacement, dedicated internet, in-house server, and failover solutions.

In practice, most of the operators who run the validator use cloud services. Cardano, the only project mentioned in the report, has a significant proportion of operators who self-host their nodes.

Among other things, the report looks in particular at which cloud service providers are used the most with respect to the number of validators and stakes.

Hosting Distribution

As we have already indicated, most operators prefer hosting services as it is a simple and reliable solution compared to self-hosting. It would be best for decentralization if most operators self-hosted their validators. Operators take the easiest route.

It is ideal if operators use a high number of hosting services and the distribution is as even as possible. Higher diversity and even distribution of nodes increase the resilience of the network.

If a hosting service provider had a serious problem that led to a service outage, Cardano nodes hosted on that service might stop operating. The Cardano network would be directly affected because if a node becomes a slot leader and is offline, it cannot mint a new block.

This works similarly for other networks. Those that have slashing can even punish a node economically for not fulfilling an expected obligation (mostly produce or validate blocks).

You can see in the figure that Cardano is doing very well, as it has the largest number of validators of all the networks mentioned. Cardano has 2,589 validators, in second place is Solana with 1,876 validators, and in third place is Avalance with 1,119 validators.

In the case of Cardana, no hosting service has a higher than 20% share. AWS has the largest representation with 16%. This statement is not true for any other network. For example, Aptos and Avalanche use 40% of AWS.

The diversity of hosting services is very important as if a large number of nodes go offline, block production can stop. This is especially true for networks that have instant transaction finality (provable finality). These networks are usually unable to finalize a block and continue with progress with the consensus if 34% of validators get offline. Cardano is made robust (using probabilistic transaction finality) and is able to produce a block even if only a small low percentage of nodes are running (say only 10%). However, the frequency of blocks would be lower (this is similar to Bitcoin's hash rate dropping).

Even if all nodes running on AWS (16%), Contabo (11%), Digital Ocean (10%), and Google Cloud (7%) went offline, Cardano was able to produce a new block roughly every 40 seconds on average.

Warning. Please take these numbers as a guide only. As we will show below, the number of nodes in cloud services is not what determines the number of blocks produced. It is the stake that is decisive.

Note that Cardano has a huge share of self-hosted validators (6.4%) and a strong representation in the OTHERS category (27%).

For Solana, the share of the OTHERS category is over 46%. This suggests that Solana is very sensitive to validator dropouts. In the past, we have seen the need to restart Solana several times, which may be related to the reliability of the hosting services.

It is not clear from the report if the Messari team actually counted the block-producer nodes or if they accidentally counted the relay nodes. In the Cardano network, block-producer nodes are hidden behind relay nodes (each block-producer node can be behind multiple relay nodes). There are approximately 3200 registered pools. The exact number can be obtained directly from the blockchain, but it is not possible to find out on which server the pool is running.

According to the stats in Cexplorer, the sum of relay nodes in the top 10 countries is more than 2800.

Let's add that some SPOs operate public relay nodes in the cloud or self-hosted and at the same time operate private relay nodes (private to protect against DDoS attacks). In fact, there may be more relay nodes than publicly available data suggest.

Stake Distribution

Similar to the number of nodes operated by hosting service providers, the representation of stakes can be tracked. Based on our suspicion that the Messari team counted relay nodes instead of pools, this information may not be correct either. However, it is likely that if an operator runs relay nodes at one provider, it will also have a block producer node there. However, it is possible that the operator has multiple relay nodes at different providers, or self-hosts a block-producer node and only has relay nodes in the cloud services.

Cardano's largest stake (amount of ADA coins) is with AWS (35%). The second largest stake is with OTHERS (27%). This includes another 139 hosting service providers. It is a relatively good result, but one has to consider that if AWS services get out of service worldwide, Cardano will temporarily produce only 70% of the blocks.

Stake representation is what determines block production, not the number of nodes in the hosting service. This means that despite a hosting service provider having, say, only 10% of the Cardano pools (nodes), if the stake of all pools together was 50%, they would produce 50% of the blocks.

The winner in this category is Solana, where the OTHERS section has a 46% share. Probably for the reasons described above. Avalanche relies on AWS for 67% and Aptos for 43%.

From the point of view of stake decentralization, national distribution is important.

Aptos would win this category as it has a relatively even representation of the top 3 countries (USA, Germany, Korea).

Cardano has a 24% stake in the US, and 27% in Germany. There are also significant stakes in Canada, Switzerland, the UK, Singapore, and the Netherlands. This figure correlates with where operators are running relay nodes (and probably pools as well) from. According to Cexplorer, it is 30% US and 25% Germany.

Let's point out one more thing that may not be obvious to everyone from the report. The total stake of pools that are operated from Germany, for example, may be made up of coins from delegates who are from different parts of the world. This means that the stake (ADA coins) is not physically located in Germany. The coins are still held by the stakers in their wallets (let's neglect the possibility of staking from centralized exchanges). Nor does the data say anything about the decentralization of the network in the context of who holds the coins, i.e. who is able to run the validators. This can be a team (holding tens of percent of the coins) or a large number of independent entities. It depends on the distribution of the coins. Furthermore, the quality of the staking must be considered. Cardano has liquid staking. So the stakers hold the ADA coins during staking. In other networks, in some cases, it is necessary to entrust coins to a third party (less decentralization, risk of abuse of power).

How expensive is it to run your own node?

How expensive it is to run your own node is mainly determined by the demands on computer resources. Cardano does very well in this area.

Avalanche has the lowest memory (RAM) requirements (16 GB). Cardano requires 24 GB. Solana requires 128 GB of memory. Cardano has the lowest disk storage requirements at only 128 GB. Other networks require 500GB to 2TB.

The Messary team seems to have made a mistake by setting the "Minimum Staking Requirement" section to 0. For a Cardano pool to produce blocks regularly in each epoch, it must have a stake in the order of millions of ADA coins. These can be delegated or provided by the operator. In the Cardano ecosystem, there are pools that are saturated and have 0 pledges (no operator stake). Maybe this is how the Messari team understood it. Anyway, a pool with no stake cannot produce blocks.

It is worth noting the relatively high requirements for the number of coins needed to run the validator on other networks. For example, 2000 AVAX (~30,000 USD), 32 ETH (~60,000 USD), or 110,000 ATOM (~1.2M USD).

It is difficult to estimate how many ADA coins of the pool operator (the pledge) are needed to achieve 70-80% saturation of the pool via delegation from stakers. There are pools that have done it with 0 pledges, some operators put in 500 ADA (160 USD), others 500,000 ADA (160,000 USD).

Conclusion

First and foremost, it is important to express a big thank you to the Messari team for including Cardano in this type of reporting. A year ago this was not the norm.

It can be relatively difficult for the teams producing these reports to understand the details of individual projects and put together relevant comparisons. The details always matter. It is important that there are independent organizations that will compare blockchain networks from different perspectives, as this gives us important information about how individual projects are doing.

I sometimes encounter the opinion that Cardano or any other network is not decentralized compared to Bitcoin because it relies on cloud services. It is important to note that Bitcoin's core infrastructure is also dependent on hosting services and large mining halls. Centralization is low. If the 2 dominant pools go offline, Bitcoin will temporarily produce only 50% of the blocks. If a large mining hall (or several at the same time) goes offline, the hash rate will immediately drop and this will slow down block production. In the past, natural disasters (floods in China) have had the effect of reducing Bitcoin's hash rate. The quality of decentralization should be judged in terms of the number of key points (single point of failures), not in terms of the number of nodes in a distributed network that does not have the ability to produce blocks.

Featured:

Related articles

Did you enjoy this article? Other great articles by the same author