how many used one of the "lightning in a box" implementations.
Not that it matters that much, but all the pre-done ones that I know of are running LND so over time C-lightning could wind up being a much smaller player.
I like the C-lightning plugins idea, but LND seems to be more widely used.
seems like a mistake
because LND's written using golang, the devs are having to deal with golang's problems
1. Difficult to find bugs where the CPU usage has frequent 100% spikes
2. High memory usage, again where spikes are sometimes difficult to explain
using golang makes it hard(er) to control that sort of thing, and it's by design, because the idea is that the programmer can focus on writing the app, not on the subtle job of also making the app talk to the operating system
c-lightning uses minimal RAM and CPU, because it's written in C, where the programmer must directly control almost everything the app does (with golang, a bunch of Google-written magic does the job for you)
So really, these small rPi nodes (and bigger servers with more than one LN node per machine) are better off ditching LND.
It's becoming more clear this will be necessary as Lightning gets more actual use; I'm running c-lightning on an rPi 3, and when the network is quiet, all is well with system resources (I've got 2GB swap space on an SSD to ameliorate the low RAM). But when there are big storms of transactions, the gossip updates push the CPUs to ~ 50%. There will be even more bigger surges in future, I expect
If LND don't fix their problems before Lightning use gets more regular (and heavier), these little rPi nodes won't be able to handle it.
Besides, LND seems to exist to sell a bunch of services to you anyway (swaps and watchtowers), c-lightning is focusing on supporting a more peer-to-peer dynamic (dual-funded channels). LND is coming to represent what critics say about Lightning: resource heavy and centralized