17

I am working on a server within my office. The server will eventually be relocated to a data center. I would like to be able to leave the server switched on in my office, which means I would like to be able to protect it from power outages or surges. In the office I only have desktop UPSs. I would like to avoid forking out for an expensive server class UPS.

I don't mind if it only has protection for a short time (Even a few minutes would be longer than any likely power drop where I live)

The UPS is APC Back-UPS ES 400 (400 VA, 240 Watts)

The server is DL 360p Gen 8 (750 watt PSUs)

  • 2
    You say PSUs, plural. Does this mean it has redundant power supplies? If so, the combined power draw can be higher than 750 watts due to the way it manages the redundancy. Also, a 1500VA UPS is not that expensive compared to the price of a broken server. – BeowulfNode42 Apr 21 '17 at 8:52
  • 3
    @BeowulfNode42 it does have redundant PSUs but I would probably only have used one, or plug two PSUs into two UPSs (I have more than one of the Back-UPS) but I see from the answers that I'm better off just not risking it. – MrVimes Apr 21 '17 at 10:05
  • 2
    Two 1500VA UPS plus the USB connection that tells the OS your power is low and to shut down gracefully will work just fine. I've done it on test servers outside of datacenters. Just understand there are other reasons running your server outside of a datacenter are sub-optimal. A single 400VA UPS will litterally smoke if your server resets. I've smoked a few of them; again, in a test environment where I don't really care that much. It is for sure a fire hazard. My APC units litterally threw sparks 2 feet. – Aaron Apr 21 '17 at 12:23
  • 1
    Silly thought: Could you plug in a UPS into a UPS to (inefficiently) sum their capacities? :O – Alexander Apr 22 '17 at 3:04
  • 2
    Don't do it; some servers require regular sinusoidal power and can be damaged if they receive the approximate square-wave kicked out by many UPSs that are designed for desktops. – SeldomNeedy Apr 22 '17 at 5:05
33

I would not plug a server with a power supply capable of drawing 750 watts into a UPS which is only rated at 240.

The issue isn't really that it's a "server" or "desktop" UPS. You're likely to trigger overload protection and drop your server even if the power input is fine.

  • 14
    Yes and no. At the end, the rating is irrelevant - real power draw is relevant. Servers somehow come with hugh supplies, even if there is no chance that is ever used (because the max draw of the hardware in it is WAY lower). I would measure draw under max load first. And no, I am not joking - I have servers in my server room that have a 900 watt supply but only a handfull of discs and a CPU / OBO with around 150 watt max - no way they ever tax it. But the case can also hold many more discs or in one case is prepared for 2 GPU units. Which are not there. – TomTom Apr 21 '17 at 10:36
  • 2
    I agree that it's more than likely the server does not draw 750 watts. The power supplies will be spec'd for the maximum load that server can draw, and then probably only when fully spec'd up. However I'd argue that, regardless, it's a fools errand to plug a server with 750watt capable power supplies into a 240watt UPS, even if you've measured it drawing 200 or less. – USD Matt Apr 21 '17 at 14:29
  • If you can, test run the server using just one power supply and measure the AC power draw. If it's consistently well under 240 watts, it should be fine to get the server through brief power interruptions. But I doubt it will be that low. – David Schwartz Apr 21 '17 at 18:47
  • The PSU is rated to deliver up to 750W. Due to inefficiency in converting AC to DC, it draws more. The specs claim 80+ Platinum certification, which implies ~89% efficiency at full load, so the max draw should be about 843W. – Nat Apr 23 '17 at 5:31
  • Also, note that during dynamic electrical events, draws can artificially spike before coming to equilibrium (inrush current), so the the draw might go over its equilibrium spec of 843W when transitioning as a transient effect (which nonetheless can cause trouble for power supplies). – Nat Apr 23 '17 at 5:37
16

Your UPS provides power when the utility mains drops, and protects from transient under and over voltage/current conditions.

Given you can plug a desktop or a server or a UPS into the same supply socket, the power is all the same, and what comes out of the UPS is the same.

However the UPS wattage and the PSU ratings are a maximum

You need to know how much power your server draws on each PSU. You can find this information with a Watt Meter ("kill-a-watt" is one brand) or an AC clamp meter, or in your case a good estimate comes from the iLO.

iLO4 power meter image from a HP DL380

The figures on the right side will show you a number in watts, and if its over 240W then your UPS will probably shut down with "OVERLOAD" if the mains goes out, or if the UPS needs to buck or boost.

Mine shows * Peak of 356 Watts which appear to be correlated with booting. * Minimum of 166 Watts * Current value of 217 Watts

I would not run this server off a UPS rated at 240 Watts but your numbers may vary.

If your maximum draw is smaller than 240W you might get 10-30 seconds power out of the UPS. Not really enough time for a safe shutdown assuming you're standing right there.

The backUPS may or may not have a serial or USB port for monitoring, so without that and the powerchute or NUT or apcupsd software, then the server will be going down hard anyway.

Finally you might be buying yourself additional problems. If the power goes out, your UPS will do something, and run itself flat very quickly. Many UPSs will not power on when mains returns, because its better to stay off than to start up with a flat battery and be vulnerable until the charge rises. So a small power blip means you have to go into work to turn the UPS on afterwards.

tl;dr In short, that UPS is probably too small, but do check first.

Your best answer is to get the server into the DC as quick as you can, and then use the iLO to do the setup and installation.

Here's a similar power meter from an older iLO2 equipped server (a HP dl380 G6) This sort of info should be found in any brand-name server that has an OOB interface (out of bandwith management, like cisco's CIMC, IBM's RSA.

enter image description here

And the same info from the DRAC6 on a Dell r610.

enter image description here

  • 1
    I'm not sure which model he has, but the BackUPS BE400 is rated for a bit over 5 minutes at full load as is the BackUPS Offices 400VA BF400C. If his load really is less than the rating on this UPS, then it should be sufficient for safe shutdown for most servers. Even without auto-shutdown it can still be useful since most power failures are short. In the past 3 years, my UPS hasn't been on battery for more than 117 seconds at once. – Johnny Apr 22 '17 at 4:21
  • @Johnny fair point - this UPS is rated at 240 Watts maximum. My HP server reports that it is using 227 Watts, that's close enough to overload to not even bother with this UPS. Hence my advice to get the server into its DC and then do all the setup via iLO. – Criggie Apr 22 '17 at 7:11
  • 1
    Agreed, he's probably going to awfully close to the power rating on the device, and if his server has a load spike while running on battery and power draw exceeds the UPS's capacity, the UPS may just shut down. Just wanted to point out that he can expect ~5 minutes of battery runtime at full UPS load. – Johnny Apr 22 '17 at 21:59
6

Back-UPS is not suitable for server protection. More or less suitable is the smart-UPS that has a way more sophisticated controller inside. The typical operation cycle for a power supply when the wall power goes out is:

  • If the remaining charge less than 30% the UPS sends a signal to the server to shut down.
  • The server correctly exits all the programs and sends a signal to the UPS "Be ready in 5 minutes" and runs the shutdown command.
  • UPS waits for 5 minutes and powers off the outlet the server is connected to.
  • UPS waits until wall power comes back and starts the battery charging until it is charged to 50%
  • UPS powers on the outlet the server is connected to
  • Server boots and sends a message to the UPS "I'm ready". Until now UPS does not power off the server under any circumstances.
  • If another outage happens during server startup, UPS sends the shutdown command immediately after the "I'm ready" signal is received.
  • Otherwise UPS continues charging
  • The end

Dumb back-UPSs can't do all those tricks.

  • Is Back-UPS suitable for anything? I've used some for network devices and they have ironically caused more power outages than the environment... ;) – Esa Jokinen Apr 21 '17 at 9:57
  • 1
    Back-UPS is good enough for PC that powered off manually by owner when UPS beeps and powered on when outage ends. That is his primary function. As far as server should operate autonomously, it should be powered on and off automatically and accordingly to some strategy. It's obvious that data on the server is more valuable than one on the personal computer. – Kondybas Apr 21 '17 at 10:02
  • 2
    @EsaJokinen I'm not sure what you're doing with your Back-UPSes, really. I run two APCs at home; one smaller Back-UPS ES unit to keep the FTTH transceiver powered through outages (which happen on a semi-regular basis for me), selected more based on Wh than VA, and a larger Back-UPS Pro unit to keep the computer powered long enough to get through some of the outages and an orderly automated shutdown if it's a longer outage. Neither has given me any significant trouble over 3+ years, and I certainly haven't had any indications that either has caused any service outages of their own. – a CVn Apr 21 '17 at 20:47
  • We have a really steady environment, all cables underground, so power outages are seldom. Every now and then I need to replace dead batteries, but having plenty of Back-UPSes that actually comes more often than the actual problem they are trying to prevent. Maybe the lack of automated self-testing is the main problem. Or just bad luck. – Esa Jokinen Apr 21 '17 at 20:53
1

If the features Kondybas mentions are not required - yes, IF a) the power rating is enough (which is not the case here) and b) there is no regulatory requirement that would forbid using desktop-grade equipment unattended 24/7 - doing so could, depending on locality, violate eg fire insurance requirements or other safety policies. Also, in case of an accident, not meeting condition a) could be interpreted against you as intentionally/negligently overloading electrical equipment...

  • And how about the data loss? Meeting the regulations is ok but even if no restrictions exists non-intellectual power supply is dangerous for data (and business). – Kondybas Apr 21 '17 at 10:41
  • 1
    Depends on what the server is doing, and how it is doing that. – rackandboneman Apr 21 '17 at 11:01
  • OK, example: a content delivery server that has not the only copy of what it is serving (if it has, where is you backup?), that you want as available as possible, that has a modern journaling filesystem, that has a BBU or FBWC system. Not only do you you want it (and the associated connection equipment) kept online to the point where your UPS is exhausted, it is extremely unlikely that there will be even a data loss from a hard shutdown. A crash resilient storage setup should be your first line of defense, UPS shutdown if wanted second, backup third. – rackandboneman Apr 21 '17 at 17:17
1

As I dont see mention of that, I will add a note. Highter end UPS will tend to be TrueOnline

Unlike a BackUPS, that when the AC input is lost it switch to the battery power, a True Online UPS always use the battery power in the output outlet.

That mean the AC input is always clear from signal range problem. The input AC always charge the battery. (AVR Trim if the signal is too hight versus AVR Boost if the signal is too low)

The battery live less, but your server will always have a good input.

There in the image you see a example of what I mean, the lower line is the normal's way for a trueonline ups, while the upper way is the bypass mode if you put the UPS in bypass mode.

enter image description here

There an example of the AVR control's I talked about;

enter image description here

  • Is this relevant for modern switching power supplies used in computers? I've never seen one that can't handle the few milliseconds of transfer time for a line-interactive or even a standby UPS. For exceptionally dirty power, I could see it being useful, but the UPS may be more picky about input power quality than a computer power supply and stay on battery all the time. – Johnny Apr 22 '17 at 4:49
  • @Johnny The goal is to protect from dirty power. As you are always on battery. What if you have a external ac generator that spike on start? etc.. while a simple ups simply protect from power outage, but I dont say he cant use a back ups, just highter end ups protect from other problem – yagmoth555 Apr 22 '17 at 12:23
  • Why the downvote? All APC SmartUPS use that technology while backups don't. If I used bad term to talk about it feel free to edit. – yagmoth555 Apr 22 '17 at 21:41
  • APC has 3 main types of UPS's -- the standby UPS (BackUPS), line interactive (BackUPS Pro, SmartUPS), and online (Symmetra, SmartUPS Online). Most people will be fine with a standby or line-interactive UPS, and it'll run cooler and use less power than an online UPS. I didn't downvote you but you probably got the downvote because you didn't answer the question. – Johnny Apr 22 '17 at 21:55
  • @Johnny Thanks for the feedback. I appreciate it. As stated in my answer, my goal was just to give more detail about the generic question, as no one talked about that. (as even someone talked about apc software for shutdown) As the op already marked a correct answer before my answer. – yagmoth555 Apr 23 '17 at 0:26
0

It all depends on the amount of power being drawn. All a server is is a computer. Just keep in mind that most servers are in racks and the power drawn and supplied as a group is much higher than an end-user's requirements would be. A Smart UPS is more likely what you're after. Having said that the power supply in the server should handle the server's specific requirements and the UPS just needs to cover that power supply's top capacity and any likely max jolt you're going to hit on that power line. In other words, if you have a 1500W power supply, then get an UPS that can provide at least that and the UPS needs to be able to handle a power spike of a certain level. Most UPS aren't designed for long-term coverage...simply enough to get done what you need to and then to gracefully shut down the machine.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.