6

To sketch the situation :

  • a bunch of servers (~15 going from 4u to 1u, some redudant PSU some single)
  • 2 UPS (1 APC I can read using apcupsd)
  • 2 PDU

So I came upon this situation, and I honestly have no idea what the best solution would be. We have currently put most of the servers (even redundant PSU) on 1 UPS (older one, no APC) Then when we started drawing too much juice from it, it started to complain and they bought another one, and changed the probably most power hungry servers (clusters) on the new UPS (APC)

I don't mind servers going offline during a power break, and I would like them to go down well before the UPS give out. Currently the servers run until they don't get any more power, then die painfully.

So the reasonable situation would be that we dedup the servers who can so that they pull 50/50 from each UPS (as far as possible). However, during a power down, one of the batteries would be finished before the other one, this would result that ALL the servers draw juice from the other UPS, probably pulling way to much power and resulting in a instant power break?

So my questions are do PSU pull 50% from each power source? And how can I (if necessary with new hardware) fix this situation. And what is the worst that can happen when you pull too much from a UPS ?

I want to protect the servers from high peaks in power and in short outages of <5 min. Anything beyond this time-frame is explainable up the ladder. Or servers are configured OK and we know they can power cycle if deemed necessary.

Canadian Luke
  • 885
  • 14
  • 41
SvennD
  • 739
  • 5
  • 18
  • What type of power outage are you trying to protect against? If the power load exceeds your battery's capacity, you have the option of scaling up. If you need runtime, that's a different question. – ewwhite Apr 16 '15 at 16:54
  • I would like to protect the servers from very short powerbreaks (seconds) and from going down hard (ups depletion). I don't mind so much doing it well before the UPS got till the end. I just don't want to introduce a possibility where -if servers refuse to shutdown- the UPS would overload and catch fire ... (or something) – SvennD Apr 16 '15 at 17:54

2 Answers2

5

A few questions to answer here...

As for the power supply distribution in dual PSU servers, the short answer is that it depends on the server manufacturer and system/BIOS configuration.

Please see: How do servers with redundant power supplies balance consumption?

With regard to using multiple UPS units, you should try to match the specs and load. I find it simpler to go with a single large UPS or multiple right-sized UPS units that serve logical groupings of equipment (e.g. Network rack, server rack).

I'm not a fan of the half UPS/half utility power approach. That introduces too many unknowns when you start mixing in equipment with single PSUs and trying to calculate loads.

Also see: Is there any value in protecting both rack power feeds with their own UPS?

Community
  • 1
ewwhite
  • 194,921
  • 91
  • 434
  • 799
  • Thx for your advice, I agre that half UPS/half uti is bad, I will however take the advice of fox, and go with a straight 5 min. power out --> shutdown. This way I can safely shutdown and don't have to be scared to break stuff. (both ups are strong enough to hold 5 minutes) – SvennD Apr 17 '15 at 07:54
2

If you don't mind the downtime, just set the servers to power off as soon as (or very soon after) the power goes out. That way you won't deplete either battery.

I personally would connect one PS of dualPS servers to the UPS (either of those) and the other one directly to power grid. That will simple up your load balancing.

Best bet at loadbalancing would be to measure the power usage of each server at expected load and then dividing the load according to UPS specs.

Fox
  • 3,887
  • 16
  • 23
  • 1
    Why would you advocate connecting one PSU to mains and another to the UPS? – ewwhite Apr 16 '15 at 16:55
  • @ewwhite Because that eliminates two of the unknowns from the equation. (Exact load behavior of dualPSU servers and load/stability once one of the UPSs goes down.) I would consider connecting one PSU to each UPS only if I were going for 1+1 UPS redundancy. – Fox Apr 16 '15 at 17:00
  • Thx for your advice, since I haven't found a way to check battery levels on 1 of the 2 ups's I can't send a kill signal at an appropriate time to the machines that are connected only through the old one + power grid. – SvennD Apr 16 '15 at 17:48
  • 3
    @SvennDhert that is why I suggested what I did in my first paragraph. You can base the shutdown on the UPS you can manage. You know the power is out and if your only goal is clean shutdown (your goal is not maximizing uptime), you can do it very soon after the power goes out (say 30-60 seconds; or 5 minutes if your UPS is large enough). You get clean shutdown and don't need management on both UPSes. Or you can do this only for the unmanaged UPS, leaving the managed UPS to do shutdown when capacity is low. – Fox Apr 16 '15 at 18:15