What size UPS do I need

[Chris Moore]

I will try to get some better metric for you when I am back at work tomorrow.

 

[MatthewSteinhoff]

using consumer drives in our NAS and that was the biggest mistake in the build

Consumer SATA drives - pulled from desktops when they are upgraded to SSD - are pretty much all we use in our file servers. They aren't as long lasting as enterprise SAS drives but their performance is every bit as good. Since we're getting the drives for, effectively, free, we don't care that they have to be replaced every couple years. Remember: the 'I' in RAID is 'Inexpensive'.

Cheers,
Matt

 

[pro lamer]

wait till the drives are in and test the draw

That's interesting to me too. I haven't bought an UPS yet and I'm wondering whether I can buy a smaller UPS instead of using the same measure as for the PSU.

In other words: should I include drives spin up current in this number:

how many amps of power is this thing pulling?

(Assuming a system with no staggered spin-up - my rig will have 7-9 hdds and the example below is told to have no staggered spin-up)

I can imagine a power outage occurs during spin up time.

Or can I assume that an UPS can resist short term spikes better than a PSU? Short term spikes = extra current draw during spin up time. Short would be 10 seconds according to this post from our power supply sizing guidance

The fine folks over at 45Drives have some pretty good info. So, here, they have a chassis with 45 drives, all spinning up "simultaneously".

In other words: how much is this extra?

when the system first spins the drives up, but it settles down fairly quick. I would say that you want to have some extra capacity above what I said

Sent from my phone

 

[Chris Moore]

That's interesting to me too. I haven't bought an UPS yet and I'm wondering whether I can buy a smaller UPS instead of using the same measure as for the PSU.

The same guidance applies for a UPS:

Proper Power Supply Sizing Guidance
https://forums.freenas.org/index.php?threads/proper-power-supply-sizing-guidance.38811/

With the exception that you need to estimate extra capacity to allow for the amount of run-time you want the UPS to be able to stay available on battery. Most of the UPS manufacturers have calculator apps on their websites, you should use one of them.

 

[danb35]

IMO, VA capacity is a nearly-useless metric for a UPS, as it only states the max power the unit can provide (which is rarely an issue)--the question is almost always the watt-hour capacity of the unit, which is a spec the manufacturers hide (if they provide it at all). If you size for the desired runtime, you'll almost certainly have more than enough VA capacity.

 

[Constantin]

That Storinator chart seems to suggest that the drives are in fact staggered, whether they intended them to be that way or not.

The spin-up suggests only 675W of power being drawn, which is about 15W per hard drive. That is much lower than the data that farmerpling2 has published... Unless the storinator switched to 2.5" drives (and I'm pretty sure it's a 3.5" rig), the data suggests to me that there is some staggering, whether they are programming for it or not.

On the other hand, the idle power requirements seem much more inline with expectations (i.e. 5.6W per drive). Additionally, I would expect the initial power draw to be be spikier than what I see here. But that's just based on observing power draws of motors in general.

My own rig suggests some staggering based on the startup whine of the 8 drives in it.

Lastly, I would size the power capacity of the UPS to match the need during normal operation. The way that UPS' are typically marketed is that a UPS that meets your runtime requirements for a orderly shutdown will typically have oodles of capacity to supply said power. My personal experience with power supplies suggests that right-sizing the output of a UPS is desirable from a run-time perspective, but I'm happy to be wrong. I'd size mine for the normal power draw, not startup, as my pool is unlikely to be written to until the system is fully up (because it's encrypted).

 

[danb35]

Using the closest current APC model to mine, their runtime chart says that at full load, it will last 3 minutes. If you're planning on having the system immediately shut down on power loss, sizing the UPS based on your running system load (I agree, startup demands really shouldn't be a factor, as you aren't going to be starting up your system on battery) could work. But if you want capacity to keep your system up for a bit, the VA rating of the UPS will be nearly meaningless--size it based on the manufacturer's runtime chart.

 

[jgreco]

That Storinator chart seems to suggest that the drives are in fact staggered, whether they intended them to be that way or not.

If they didn't design a custom-built bit of hardware to do staggered drive spinup, they're idiot-tards. There's a lot of issues related to this in the data center, you can't always rely on A/B to save you. Have a full rack of 11 of those things spin up and do the load calcs. Fun!

Also, as the number of drives increases, the chances that there will be some significant deviation in the exact timing of load spikes increases dramatically, because even when drives are all powered simultaneously, they do not typically spin at the exact same moment. This does help reduce the scale of the problem.

My PSU sizing guide is aimed at calculating the maximum load based on the possibilities that this stuff might not happen, because it is aimed at hobbyist-class systems built from parts that probably don't do staggered, and have maybe half a dozen drives. At those sizes, the pragmatic thing to do is to just size the power supply appropriately. I actually published the PSU guide because there were some people wandering around here, pretending to be experts, arguing that you should size your PSU as small as you possibly can. Since there is a nontrivial risk of expensive equipment damage, and home users probably don't have a budget to replace thousands of dollars of gear if their PSU smokes their system, well that's why I wanted to provide a "safe" guide.

The design of these things for the data center is a different problem. I've wrestled that one for years - I build systems professionally, and did stuff like 4U/24-drives before most others did. In general the large storage gear mitigates this in a variety of ways, and this can include A/B power, staggered chassis spinup (multiple units in a rack), staggered drive spinup guaranteed through a variety of techniques, limiting drives to ones known to behave in certain ways, etc.

Using the closest current APC model to mine, their runtime chart says that at full load, it will last 3 minutes. If you're planning on having the system immediately shut down on power loss, sizing the UPS based on your running system load (I agree, startup demands really shouldn't be a factor, as you aren't going to be starting up your system on battery) could work. But if you want capacity to keep your system up for a bit, the VA rating of the UPS will be nearly meaningless--size it based on the manufacturer's runtime chart.

And yes, size based on runtime calcs (chart or otherwise). I know some guys out west where the power isn't super-reliable who are using large UPS's primarily for the extended runtime for their lowish-watt applications. Three minutes runtime (mentioned somewhere above) is particularly troubling because it is probably three minutes with a brand new set of batteries, and might not sustain that for even two minutes once the batteries are a few years old. It also implies that the UPS is pretty small. You are better off with a UPS that's somewhat larger than what you think you need.

Also try to remember that with the way autoreclosers work, power is likely to go out, come on, go out, come on, and go out again for a hard fault. This will usually happen all within 60 seconds. Most of the time, if power is unstable or out for more than 60 seconds, expect that there may be a crew that needs to be dispatched to remediate. This should help inform your decision-making as to what your goal for a power outage is. A five-minute runtime is a good idea, in that it provides enough time for you to get into a hard power failure, then decide to shut down, probably at the two minute mark, and then be safely shut off by three minutes, which gives some safety margin for old batteries etc.

 

[MatthewSteinhoff]

if power is unstable or out for more than 60 seconds, expect that there may be a crew that needs to be dispatched to remediate.

In many cases, the UPS just needs the hold the load long enough for the generator to spin up. Our worst case scenario is three or four, two-minute outages in a row that drain the UPS while staying below three-minute point where the generator takes over. Hopefully, Operations is paying attention and hits the manual start button before all hell breaks loose.

With 60 bays and related infrastructure, @oasismin, you may be exceeding the ability of random people on the internet to size your UPS. We don't know enough about your environment and needs to provide a thorough answer.

Cheers,
Matt

 

[jgreco]

In many cases, the UPS just needs the hold the load long enough for the generator to spin up.

"In many cases." This definitely does not include the case where a generator fails to spin and people go into a panic.

BTDT.

It's good to size it for runtime above and beyond the bare minimum.

 

[pro lamer]

Just realized I had missed this thread is about 60 drives. I didn't want to try to hijack. (Anyway I failed to hijack ;) ). Thanks for your help here. I'll post another thread some day regarding UPSes for small rigs.

Sent from my phone