camosoul
Well-known member
People who run a lot of MNs tend not to talk for understandable reasons.
I thought I'd report some rudimentary specs on resource usage that I've observed, for the benefit of those who don't own their own hypervisors and use rented VPSes instead of their own actual hardware.
I find my 0.12.1.3 MNs don't care for more than 1024 of RAM in strict MN operation.
I give them 4096, and they don't use it. This statement doesn't account for KSM, which provides an ever increasing benefit for those running multiple nodes on the same host. More RAM becomes useful when running OS updates. But, I don't find it mandatory.
I give them 4 cores, and they don't care. I barely see more than 3% CPU usage, ever, and that's on only one thread. Not including initial wallet generation and chain pull, of course... I use keypool=0 in my MN's dash.conf
Some of this refers to the capabilities of the underlying host hardware, which is totally badass and way over the top overkill.
I find the only real bottleneck is storage. Hard Drives are a bottleneck on a single machine. Run 40 VPSes off of a single piece of spinning rust, and latencies really, really pile up... There is a certain advantage in running multiple cheaper/older pizza boxes instead of one big one. More hard drives = faster. It's hard to rent a "badass" machine with a really big SSD. RAID rust doesn't really help under these extreme IO loads. It helps, but you get better performance from multiple whole machines acting independently. You end up having to cohost or spend a ton more with the single machine option. A handful of dirt-cheap older machines works just as well and gives you more hard redundancy on top of it. You're paying more for the electricity, Us, and switch hole, than the hardware at that point... Cohosting sucks. You're responsible for everything, and you pay a premium on top of it. Better to just get a stack of older machines and let the maintenance be someone else's problem. You're really not paying for the hardware at that point anyway, just the service of it not being your problem, which is absolutely worth it especially on old hardware.
Even if you run only two or three MNs, the cost of 1GB RAM VPSes makes it worth running your own Proxmox on an older 8GB pizza box.
You might be "consolidating" your own nodes, but in the grand scheme, this actually diversifies the network by pulling out of massive VPS service providers hosting hundreds or thousands of MNs under a single entity. It's also cheaper.
This would be an ideal place for cheap dedi hosts with older hardware, which also accept crypto payment, to advertise their services. In fact, it would be welcomed... If you don't accept crypto, we can totally set you up...
I thought I'd report some rudimentary specs on resource usage that I've observed, for the benefit of those who don't own their own hypervisors and use rented VPSes instead of their own actual hardware.
I find my 0.12.1.3 MNs don't care for more than 1024 of RAM in strict MN operation.
I give them 4096, and they don't use it. This statement doesn't account for KSM, which provides an ever increasing benefit for those running multiple nodes on the same host. More RAM becomes useful when running OS updates. But, I don't find it mandatory.
I give them 4 cores, and they don't care. I barely see more than 3% CPU usage, ever, and that's on only one thread. Not including initial wallet generation and chain pull, of course... I use keypool=0 in my MN's dash.conf
Some of this refers to the capabilities of the underlying host hardware, which is totally badass and way over the top overkill.
I find the only real bottleneck is storage. Hard Drives are a bottleneck on a single machine. Run 40 VPSes off of a single piece of spinning rust, and latencies really, really pile up... There is a certain advantage in running multiple cheaper/older pizza boxes instead of one big one. More hard drives = faster. It's hard to rent a "badass" machine with a really big SSD. RAID rust doesn't really help under these extreme IO loads. It helps, but you get better performance from multiple whole machines acting independently. You end up having to cohost or spend a ton more with the single machine option. A handful of dirt-cheap older machines works just as well and gives you more hard redundancy on top of it. You're paying more for the electricity, Us, and switch hole, than the hardware at that point... Cohosting sucks. You're responsible for everything, and you pay a premium on top of it. Better to just get a stack of older machines and let the maintenance be someone else's problem. You're really not paying for the hardware at that point anyway, just the service of it not being your problem, which is absolutely worth it especially on old hardware.
Even if you run only two or three MNs, the cost of 1GB RAM VPSes makes it worth running your own Proxmox on an older 8GB pizza box.
You might be "consolidating" your own nodes, but in the grand scheme, this actually diversifies the network by pulling out of massive VPS service providers hosting hundreds or thousands of MNs under a single entity. It's also cheaper.
This would be an ideal place for cheap dedi hosts with older hardware, which also accept crypto payment, to advertise their services. In fact, it would be welcomed... If you don't accept crypto, we can totally set you up...
Last edited: