A deep dive into Polkadot's DHT Network Structure
The ProbeLab team is producing Network Health Weekly Reports for the Polkadot P2P network since early 2024. Funded by the Polkadot Treasury [Polkassembly submission], the team has taken up the task of identifying the most important metrics for the P2P layer of Polkadot’s blockchain network and adapted its suite of tools to monitor the network.
In this post, we give an overview of the most important metrics and critical findings from the last six months of publishing weekly reports and investigating the Polkadot P2P network.
We use the Nebula crawler to gather information about the Polkadot P2P network. Our crawler runs every 2 hours and it takes around 3 mins to complete a crawl, resulting in a total of 84 crawls per week. This results to almost 3M visits, on average, with a visit being defined as “an attempt to dial or connect to a peer”. Below is a high-level summary of our findings. Further explanations on definitions can be found at: https://probelab.io/polkadot/dht/2025-02/#general-information
9,97113,3858,919The reports show the agent distribution throughout the duration of a week over time [link] (shown also below) and as a bar chart [link]. The results are split between Validator and Non-Validator nodes. As expected “Parity Polkadot” is the dominating agent in the network.
It’s worth noting the discrepancy between the total number of nodes seen in the network, reported further up (Network Size of close to 9k nodes) and the number of nodes seen over time, which is closer to 1.3k. This is due to a large number of nodes being unreachable, or offline, as is also proven by the very large number of timeouts seen in our “Connection Errors” plot [link].
We reported this in the past [link] and we welcome explanations as to why this might be the case, or if further investigation is needed to improve connectivity performance.
Our weekly reports are also revealing the agent version distribution of each agent, again over time [link] and as a bar chart [link]. This is very useful in order to monitor new version uptake, as well as dependencies on older versions of the protocol.
The Polkadot network seems to have a nice distribution of node deployments across North American and European states, and a 5% node share deployed in Asia [link]. It is notable that 13% of nodes have a multi-region deployment.
In terms of dependency on centralised cloud provider infrastructure, the Polkadot P2P network presents a healthy balance with more than 40% of nodes being deployed on non-cloud/custom infra [link]. The remainder of nodes are also presenting a nice balance of deployments between different providers with no single cloud provider having more than 10% of nodes.
The reports are also showing stats related to the protocols observed being supported by network nodes [link], the number of connection errors [link] and crawl errors [link] together with the error description.
The Polkadot P2P network presents a healthy state with nodes being stable overall and having a good balance between cloud- and non-cloud deployments, as well as a good geographic distribution. It remains concerning that there are a lot more nodes “seen” in the network than those that accept our connection attempts. This, however, requires a more in-depth study to narrow down on the exact problem and come up with the right optimisation.
The links and plots provided in this post are from Week 02 of 2025. Reports for the entirety of 2024 can be found at: https://probelab.io. Reports are published weekly on Mondays at https://probelab.io, so make sure to follow along, or subscribe to receive updates: https://www.probelab.network/contact.
If you have metrics in mind that you would like to see included in the weekly reports, or ways in which our methodology can be improved, please reach out to us via the details in the above link.
