15

I've done some looking and have not found a satisfactory answer to this question.

When is it necessary or suggested to use an independent UPS (on it's own mains branch circuit) for each power feed to a rack? To elaborate; when using servers with redundant power supplies, it's typically recommended that each '1' power supply goes to branch circuit 'A', and each '2' power supply is fed by branch circuit 'B', but do you need a UPS for both 'A' and 'B' feeds, or just one?

Background: I've been tasked with planning an infrastructure upgrade to our internal-use servers. We are a manufacturing company and much of our equipment and custom software relies on having reliable and online servers locally in our office. We currently have two server rooms with one rack in each, that are on opposite ends of a large building (for redundancy against fire or accident etc). They each have a UPS that is fed from circuit 'A' and the equipment we have that does have multiple PSUs has both connected to this UPS. So if the UPS were to die... ploop, there goes the servers.

My personal theory is that having PSU1 of each server connected to a UPS, and PSU2 of each server connected to a PDU that is fed directly from mains, would provide a good mix of reliability and cost. But my supervisor thinks that we should run two UPS units, one for each feed.

If this question is too vague please let me know and I will try to revise it.

cathode
  • 328
  • 2
  • 11
  • UPDATE: We purchased an 8kVA unit, exactly the same as the one ewwhite shows below, although we didn't get the auxiliary battery pack. It was much simpler to connect our equipment because we no longer had to care about dividing up the servers or power supplies. We purchased two CyberPower horizontal 12-outlet PDUs and have everything connected to those. – cathode Oct 13 '14 at 20:43
  • IMPORTANT UPDATE - Keep in mind that when a dual-supply server loses power to one supply, the current draw by the other typically doubles. This can be critical if you put one leg on city (mains) power and the other on a UPS. Thus, if you don’t plan ahead by leaving lots of extra capacity on the UPS, loss of mains power could cause the UPS to exceed its max load and either shut down or bypass-transfer to (the dead) mains power. In that scenario, a power failure renders the UPS suddenly useless. Been there... A similar situation can occur if you use two smaller UPS units instead of one big one p – user506079 Jan 19 '19 at 17:50

5 Answers5

9

Answers to this may vary. But I actually work in warehousing and logistics environments, and I'll typically plan new server rooms to have a high capacity central UPS (6kVA or more), usually an online (double-conversion) model. Devices of this tier have UPS management interfaces, so it makes sense to tie those in to SNMP and email alerts. I like to back everything up with a generator at the facility. If a generator solution is not possible, I add up to 4-hours of battery runtime behind the online UPS.

At the rack level, I'll usually run one PDU per rack. I'm not as concerned about losing a leg of power at this scale. It's totally about losing a redundant power supply on a server or device. Plus remember, not every device has dual power supplies. Many of the networking devices we see most often (firewalls, switches) can't accommodate dual PSUs, so those are definitely things you want on the same PDU and UPS feed.

A/B feeds are really for the realm of co-location facilities, where you actually have guarantees that power is being provided via diverse circuits. If you can't guarantee that, I don't think you're protecting much by having one branch running off of utility power and another on the UPS. UPS failure shouldn't be isn't that common.

8kVA APC SmartUPS R/T with additional batteries in my client's facility. It's never gone down in THREE years, despite several acts of God. enter image description here

ewwhite
  • 194,921
  • 91
  • 434
  • 799
  • Seen botched UPS maintenance (personnel don't do it often so if rushed often do it wrong) take servers down for extended periods. The other common problem on small deployments is lack of monitoring - it runs fine for 3+ years so nobody keeps tabs on it. Eventually the batteries fail so anytime the UPS kicks in it can't handle the load and shuts off taking everything relying on it down. – Brian Jul 30 '14 at 23:10
  • I'm partial to the Symmetra's myself: http://i.imgur.com/6YFLkre.jpg - :) – TheCleaner Jul 31 '14 at 03:03
  • Thank you everyone for the helpful information. The setup that ewwhite describes seems to be the kind of thing we need. – cathode Aug 01 '14 at 21:26
6

That is a tradeoff of cost and complexity for additional uptime.

For gear running in an office data center or closet I've always just used UPS for one half of the power.

In a proper data center (Equinix) we get dual path redundant power each with its own UPS.

One other advantage to running both on UPS is that you get a better ability to monitor and track current draw. This can be helpful in planning.

In short, if the money is there to go dual UPS, it is a better design in terms of fault-tolerance and uptime.

dmourati
  • 24,720
  • 2
  • 40
  • 69
1

I do that as well - one PSU off a UPS and one PSU off a PDU. Experience shows UPS are a source of failure and having one power source bypassing them can improve uptime. Make sure the PDU side has a line conditioner / AVR or one is added to it.

The main drawback is you lose the ability to power cycle the server using out of band management of the UPS. That can be a plus however as somebody making the UPS cycle power won't take down your server accidentally like during a battery run time test. It is rare these days to have a server so locked up it can't be rebooted using management tools like IPMI.

Brian
  • 3,386
  • 17
  • 16
0

Having noted several installations and from a bean-counter perspective.

trade-offs are in terms of Uptime, Capacity and cabling. You could distribute the backup and not have enough capacity or have capacity but spend on cabling.

I think that you could have IPMI/ILO tools for Power management and also look at installing NUT

Michael Hampton
  • 237,123
  • 42
  • 477
  • 940
0

I know this is an older thread now but I thought I'd add my experience.

In 2 power circuit environment it does make sense to connect two UPSs (1 for each power feed). If you use Network management cards in them and configure them properly, you can enable the software to shutdown the server only in cases where both UPSs have lost power. This protects you against false positives such as single power supply failure and UPS failure.

In the case where it is a single circuit the benefit is less although it still protects against a single UPS failure - which does happen from time to time and will not shutdown unnecessarily for it either.

I personally used this configuration with APC UPSs however I understand this is a fairly common configuration and is available with HP UPS and other major vendors.