Hi! I think it would be nice to collect some statistics on low-powered nodes to have some data on what’s the lowest reasonable requirements for a node.
I’ll start with my Thinkpad X61: Intel Core 2 Duo T7300 (nominal CPU Mark rating: 717, but SMP is disabled and the CPU is downclocked from 2GHz to 800MHz, so I’d estimate the current rating at maybe 200), 4 GB RAM, internal 1TB drive and an external 6TB CMR drive.
The node on the internal drive is long full and don’t get much traffic, but the external one is about half-full and gets around 500kB/s to 1MB/s for few days now. CPU usage is at about 15% to 20%, memory at 500MB used (both nodes + OS + streamripper running). So I suspect an even slower CPU would be enough as well—but then, C2D were one of the first amd64 Intel CPUs, so much older CPUs wouldn’t be able to run Docker.
As I wrote before, about 25W at the wall.
What are your stats? Please post only low-power devices (Xeons are not low-power ;-))
I have 4 PCs I5, I3, Pentium G series, xeon E31320 2xUPS devises 25 nodes- All consumption together 320W this is max 13W per node. All windows 10 setud. Not made any changes in CPU as all modern CPU making clock stepdown automaticly when no need to make hard tasks.
I have an Odroid HC2 (running an arm octa-core cpu) with an 8TB hard drive that’s almost halfway full (3.6TB out of the 7.5TB I assigned).
Success rate is at 99.6% since the last update so I don’t seem to miss out on many uploads even on such a low power cpu.
Internet connection is 35ish Mbps download and 9 Mbps upload.
My setup doesn’t really fall in the low wattage range… but here it is…
SGC = 280 / 24 = 11.6TBwatts - PS i do have 60TB in the system… so it’s more like 4.7 TBwatts
but only have 24tb dedicated to the storagenode because of redundancy and because i had to pull two 6tb sas drives out of my storagenode pool…
My system should run around 3W, add 8.8W to that to take the 8TB seagate into account and it should be at 12W MAX.
I would expect it to run more around 10W given that I’ve never seen more than 25% CPU usage (even that time I got 170GB of ingress in a day).
So that gives us 12/8 = 1.5 TBwatts which isn’t bad at all.
However the limitation is that I can only run one SATA drive and eventually a second one via USB but it’s a USB2 port…
I’ll give it a go sometime though, might try to hookup a cheap SMR drive to see how it behaves. I’ll setup a few other nodes before so that I can spread the load across them and not overwhelm the SMR drive.
I think a more sensible metric would be TB/kW. The higher the number, the more efficient the system.
e.g. @Vadim 120TB/0.32kW = 375TB/kW @Toyoo 7TB/0.025kW= 280TB/kW @kevink 7TB/0.06kW = 117TB/kW
i did consider using kW so that it was easy to convert into kWh, which is an option…
i don’t like that the metric in your suggestion doesn’t have any real world number from the storagenode/nodes in question…
the TBKwmonth the maybe… so like (watts8760)/12 = TBkWMon lol
terrible acronym tho… sort of makes sense tho… 8760 is 24*365 so the avg month become a more accurate number and one doesn’t have to remember stupid numbers to find it if needed.
and watts to kwh is simply adding hour behind it… and one is off to the races
you got what 16tb of total space right… so it would be 16tb … maybe one should switch the acronym around…
make it kWTBmon that way its sorted in the same direction as one would line it up when calculating the numbers…
=== kevink ===
60 Wh * 8760 = 525.6 kWh per year / 12 = 43.8 kWh / 16TB = 2.7 kWh pr month pr TB capacity…
(damn that is a nice easy to remember number actually 876 lol)
the math of it is a bit of work even tho straight forward… but one get’s a much more useful number, because one can directly use it with the electrical costs to calculate cost pr TB capacity of one’s setup.
my acronym or whatever its called is terrible… might just drop idea… but i think the number or ratio that comes out as the result should be somehow applicable in the real world instead of just a ratio that defines efficiency…
this way it still defines efficiency and also add an easy conversion into relevant overhead costs
the problem with kWh/TB is that it is an uncommon metric. If you want a metric for efficiency (which is the topic of this thread) you have to go by efficiency=output/input.
So in our case the input is the power (kW) and the output would be the TB used by the storagenode. Therefore efficiency=TB/kW.
(You could use kWh instead of kW but where’s the point of adding a time constant to it when your TB used is only a snaphot. So if you want to use kWh you have to use TBh too and then the time constant cancels itself out.)
I wasn’t exactly looking for a ranking, more a list of ideas of how to run low-power nodes. But I like it!
My Thinkpad’s external drive is connected using USB2 (no USB3 on that old hardware) and it’s fine so far with 3TB of data. I wouldn’t worry that much. On the other side, USB drives need additional power brick, and these tend to not be as efficient as integrated power supplies…
Might as well take this calculation to its logical conclusion. I have about 16TB stored, you make about 3.50USD per TB per month. My node uses about 100W=0.1kW.
(16TB*3.50USD)/730h/0.1kW = 0.77USD/kWh
I still have free storage though, so that number will go up. And also this device was already an always on device. I’m pretty sure it would have used most of that 100W even without Storj. Maybe Storj has added 20W at most because the HDDs are more active. So realistically you might have to x5 that number in my case. But for the total setup it would be 0.77USD/kWh.
I thought about that metric too but the goal here was to compare low-power node setups (hardware) while introducing money makes it more like an earning estimator with the value/TB possibly changing a lot, while the power consumption of a setup will stay the same and is therefore easier to use for comparison.
Also the introduction of a time variable doesn’t work out:
why not switch it around then and make kW the variable, if we are looking at efficiency… the dollar value is relative to electricity prices in each individual country so thats difficult to use…
however one might have some kind of restriction on power utilization… like say having a small solar setup from which one knows the avg excess power and want to pick the largest possible setup without blowing the power budget…
from how it’s setup now one would need to do math to figure that out… if the variable was the kW then one would basically just have to sort them…
and to reach a full kW one would simply have to multiply but really why would one want the kW… well to figure out the power cost in which one is already doing math already… so makes sense to bundle those two calculations.
