Some of my 2017 mini PCs have dynamically adjusted up over 450 nodes in 8gb ram. This is with standard 5k 2.5" laptop drives for data and an m.2. SSD for apps/logs/swap.
Less that 0.5 load per core, only 2gb of swap is used.
So, I’m surprised you hit the buffers on the above. I’m sure my boxes would feel the pain as.load ramped up, but that is why I have them adjust themselves (I can’t predict the load).
The boxes only have about 2.25TB, so would struggle to fully host more than around 60 or 70 half full nodes. With a bit of CPU load, it would also be a lot fewer than 450. I may also find my bandwidth gets gobbled with so many active.
The problem is, nodes are so easy to run right now, because they do so little. It was always going to be a game of optimising systems for nodes. If you don’t play the game, you get a dwindling share of the rewards, so game theory dictates this result.
Emissions are just too overwhelming vs data upload fees and/or effort to store limited data. It needs rebalancing.
I didn’t intend for my nodes to use swap at all tbh, but linux shuffles the resources over time to make space.
Given those machines have 8gb ram, 2gb swap isn’t too crazy though. Having 8gb ram and 8gb swap would probably be pretty awful. Certainly, on my boxes, linux never tries to use any more, even if CPU time is available.
Yes not yours, the increase swap to make up for memory barriers did the rounds on discord for a while.
I guess it is widely abused and causing trouble. I turned swap off completely on all machines I run lately as I would rather they get killed than harm the network.
can see it goes all over the place but i wouldn’t know if I did not have NTracking graphing it all.
node start times were 2 min interval after i noticed the hard drive io times i droped the node count and this is with only 50% ram consumed so i was not using swap.
same machine today running 1.5k nodes pretty much idle
That’s what I thought until @neo made the point that that would often prevent casual home users, or people from less-connected geographical areas from participating in the network.
What if the heuristic was kicking misbehaving nodes, but only deprioritizing underperforming nodes?
