The bottleneck isn’t the CPU on anything resembling a modern system. On mine, the storagenode process wants 5% CPU if it’s busy, and that’s on a 4-core 2GHz Xeon.
More interesting is the disk speed and the network bandwidth. You don’t earn money when the client already has their file when you get around to serving your chunk AFAIK. Thus you should have enough RAM to keep the “hot” parts of the database in core and a quick disk to load the files.
Thus I would allocate 200 MHz of CPU plus 200 MByte of RAM for each TByte served. External bandwidth can’t be high enough. My system has reasonably fast disks and 10GBit all the way, but I still get lots of cancelled uploads – a tenth of a second after they’re started.
I think this is due to the change they made to uplinks, they close the connection immediately to remove some overhead. These messages should get changed with upcoming updates.
I’m on RPi 4B and its poor little ARM processor may become the bottleneck one day considering queries have already pushed it to 130+% out of 400% (it has 4 cores), and we’re only in the early days of Storj v3
Ha ha. I feel like I’m not playing in the same league, lol.
It does consume very little power though: 260wh per day