~/notes / node operation
What Running Infrastructure Teaches You
Ten years of operating nodes has changed what I think matters about these systems, mostly in the direction of caring less about the parts people argue about.
I have run nodes since 2016, across several networks, through upgrades, outages and one chain split. The experience has shifted what I consider important.
What turned out to matter
Client diversity. A network where one implementation dominates has a single point of failure that no amount of node distribution addresses. A bug in a supermajority client is a network-wide event.
This is the risk I would rank first and it is the one least discussed outside the operator community.
Sync time and storage growth. These determine who can participate. A network whose requirements grow faster than consumer hardware is becoming more centralised continuously, quietly, with no announcement.
Upgrade coordination. How a network ships changes says more about its governance than any published document. Networks that ship carefully, with testnet discipline and honest delays, are demonstrating something.
Operational simplicity. Software that runs unattended for months, fails clearly when it fails, and recovers cleanly is worth more than software with better benchmarks.
What turned out to matter less
Throughput figures. I have never once cared about a chain’s advertised transactions per second while operating it. What I care about is whether my node keeps up, which is a different question.
Consensus mechanism arguments. Both families work. The differences that affect me as an operator are mundane: key management under proof of stake, hardware and power under proof of work.
Roadmaps. I have watched a lot of announced features not arrive. What ships is what is in client release notes.
What surprised me
How much of the system is people. Upgrade coordination, incident response, the decision to restart a halted chain. The protocol handles the parts that can be automated and the rest is a social process that mostly works and is invisible until it does not.
How rarely consensus fails. In a decade, the major networks have had remarkably few consensus incidents relative to the complexity involved. That is a genuine engineering achievement and it is taken for granted.
How much of the value is in verification rather than in features. The thing I actually use my node for, daily, is knowing that my balance comes from data I checked rather than from an endpoint I trusted.
What I would tell someone considering it
Run one if you hold enough that verifying rather than asking is worth some effort, and if you find the machinery interesting. Those are the two good reasons.
Do not run one expecting rewards, influence or faster transactions. It provides none of those.
For the portion of holdings that lives at a venue, none of this applies and a different set of questions does: regulation, segregation, withdrawal reliability. Platforms publishing that, such as a platform with real on-chain withdrawals, are answering the questions that matter for that tier, which are not the same questions at all.
# Corrections and technical nitpicks are welcome. Send them over. They get published with the fix.