I’m getting off-topic: but I’m surprised the 3.5"-tray-with-baby-SBC layout never really took off (like the Odroid HC2).
I loved the idea of just stacking them up as you filled Storj nodes… because for the price you pay for 20TB+ HDDs an extra $50 for the SBC hardware to run a node didn’t seem so bad.
Yes, I was thinking the same thing when I saw it. I know some use those, but I don’t know the limitations. Maybe just the limited RAM is the problem. Let’s move this conversation to the Rig topic.
I will look at newer models because this this obsolete. Maybe we can come up with a modern solution and test it in practice. I always look for cheap, but reliable solutions to expand.
I was really hoping Argon40 would make an EON case for the Pi5.
With its PCI interface it would probably perform fairly well and it would be a sweet way of having 3 or 4 HDDs running on a self contained form factor.
Alas, it is currently not in their plans…
Orodroid H4 Ultra is a beast! I didn’t knew about it when I built the N100 system.
It outperforms N100, supports 48GB DDR5, and comes with M.2 + 4xSATA3 + power connectors.
The SATA bandwidth is limited to 8Gbps, but I believe you won’t need more for 4 nodes.
It has 2 Ethernet ports.
I think you can build a pretty good 4 node system around it.
It out of the scope of the topic, 1 board - 1 node, but it’s a nice option.
I was referring to setups where there is just an HDD attached to the pi, or anything else. it will be game over once node requires over 200-ish iops HDD can provide. Adding SSD cache would help address this, but it’s not longer “hdd on pi” setup.
I thought products like ZimaBlade at $64 would be ideal candidates. Two SATA ports to start with — one for HDD, one for SSD cache, and 4-lanes PCIE 2.0 port, to plug in the HBA instead and have a proper storage solution.
That’s on expensive overkill though.
Well, yes.
But this isn’t likely to happen for a few years yet, is it?
Or is something coming that I missed?
Ooooh, I’d never heard of it!
Seems like a nifty bit of kit (although I have a distaste for Intel Atom CPUs).
Might get one to play with 
EDIT: Come to think of it, I can buy a used Dell Poweredge R320 with a Xeon CPU for not a lot more money on eBay so I’ll probably go with that for future nodes (space allowing) 
Recently storj did a test from salt lake satellite sending slightly increased traffic, in the ballpark of 50-70Mpbs. A lot of nodes were killed when they ran out of ram because disk would not keep up, a few people complained very loudly on the forum (seriously? Do you want more traffic or not?!). The test was in preparation for/ in anticipation of a customer getting onboarded, at least that’s what I remember reading.
So this might happen sooner than later. Which is a great thing. Just not for a potato single hdd appliances.
Nothing wrong with atom, C3000 series are pretty good.
This was exactly my thought process to. Huh, nice, and proceeded to buy old supermicro server from local hardware recycler
If this is true it indicates a poorly designed software in the first place. Nodes should just stop accepting new uploads when the unprocessed queue grows too much.
Not necessarily. If you allow nodes to temporary fail to send or receive data the redundancy would need to be reviewed, increased, and ultimately it will be more expensive to store the data. And what for? To allow potato nodes on the network? Why? Let them die off, and the quality of the network will improve.
Node shall handle anything thrown at it at the stated egress and ingress limits. If it doesn’t — it’s a bad node.
This is what they told us…
But I agreee, the future is enterprise hardware. I guess one day they will make it select only.
“Help, help, I’m having money thrown at me!..”
“…so much money! It’s overwhelming!..”
“…it feels like I’m being bludgeoned to death with money! Somebody please make it stop!..”
“…I don’t know how much more storage I can be paid for… or transfer I can be paid for… nobody should have to endure this!”