End-user nodes usually check it 3-4 times a day, so the idea is to have node spinning up just once the user comes online to use the node.
Optimization was done on start-up, spinning up in 1s, connecting to peers in a couple and be ready in a short amount of time.
Always-on is an illusion.
Core-lightning architecture:
- Main daemon: lightningd
- Sub-daemons: hsmd, gossipd, connectd, etc.
- Plugins: arbitrary scripts to customize the node.
This follows UNIX philosophy: "Do one thing and do it well".
Blockstream runs many nodes, with no control on any of the funds of the user, nor of his keys:
It's a VLS signer on the user side that talks to hsmd through a proxy to use the node.
On a daily basis, Blockstream consumes around 72 TiB/day of data from the main chain to keep all the nodes up-to-date.
This is done on just two bitcoind instances!
