After the December 2010 High-end and Mid-range Buyer’s guides, I received a lot of questions asking how redundant power supplies, and why I would recommend them over high-end enthusiast power supplies in many cases. This article will cover an overview of basic redundant power supplies, not n+1 units (that will be another piece soon).
The use cases for redundant power supplies generally involve:
1. Business applications where downtime impacts either user or customer access to information and applications.
2. Co-located servers where a technician may be hours or a day away from servicing an equipment failure. Usually this scenario involves some sort of business application described in 1.
3. Mission-critical applications where nodes cannot fail. For example, scientific application infrastructures that are not fault tolerant need to have the ability to be serviced mid-run.
Unlike disk failure, power failure causes an entire system to be knocked completely offline. That is why modern datacenters spend so much time and money with building redundant power supplies, investing in power feeds from different grids, backup generators, and large uninterruptible power supply (UPS) units. Redundant power supplies are built to work into the redundant power systems in datacenters and ensure that power gets delivered to all functional server components.
Anatomy of a Redundant Power Supply
For this article I am using a fairly standard 600w EMACS MRM-6600P RPS redundant power supply that is meant for dual CPU servers with two 8-pin CPU connectors. Zippy-EMACS is a fairly well known supplier of quality server power supplies with a much lower profile in the enthusiast power supply segment.
The first thing one will notice is that the EMACS power supply is a lot larger than a non-redundant power supply. Of course, with these power supplies, a lot more is going on than in non-redundant units. It should be noted above that the standard PSU is a 550w unit while the redundant unit is a 600w unit, but it is a decent representation.
Above, one can see the PSU carrier. Here the dual (redundant) PSUs are housed. Looking inside one can see the PCB interface between the power supplies and the backplane cables.
Each PSU is installed into the carrier and pulling the PSU does not interrupt the flow of power, neither does removing the power cable to one PSU.
One will notice that the AC plugs are not in the rear of the unit but rather on extended cables. I purchased this PSU to build a server with front-mounted power plugs, which is the reason for this configuration. Many/ most power supplies have power plugs in the rear of the units.
Redundant PSUs also provide one feature that is often overlooked. One can actually use redundant PSUs while changing UPS units plugging one unit into the new source before unplugging the old UPS, keeping the server powered throughout the process.
That underscores two common configurations of power sources for redundant PSUs. One option is that users can plug both redundant PSUs into the same power source (often a UPS.) Second, one can connect the PSUs to different power sources (different grids, diesel generators, and UPS units.) One major consideration in the second scenario is that each source of power should be conditioned similarly.
One important note is that the mounting of redundant PSUs is generally different than non-redundant units. For example, the Norco RPC-4220 ships with a standard ATX mounting bracket that mounts using four screws into the chassis. To install a redundant PSU, one needs to remove this bracket and screw the redundant PSU’s carrier into the case directly. Removing this bracket is important because with the bracket in place, one cannot pull the PSUs from the chassis. This is a reason that some consumer cases will have issues accepting redundant PSUs without modification.
Hopefully this overview shows why redundant PSUs are important features of many storage servers, and how they work. One major negative of redundant power supplies is that they cost a LOT more than standard PSUs. It is not uncommon for a 600w redundant PSU to cost $400-500 while a non-redundant 600w PSU costs $50-100. Clearly the market supports a high premium for redundancy. With many power supplies lasting 3-5 years or more, there is a good chance that one will never need the redundancy provided by a redundant power supply. On the other hand, for many businesses, the few times when the redundancy does come into play easily pay for the purchasing cost differential.
Great writeup on redundant PSUs. One question I have about them deals with AC power draw. Intuitively, in a redundant PSU, both modules have to be delivering at least some amount of DC to the system at all times—you can’t have one module “sleeping”. But exactly how much power does each module draw? Is the power usage exactly 2x what it would be with a single PSU? Or does each module pull about 1/2 what a single PSU would (resulting in total power draw that is about the same)?
The last method; they load share. In total they draw slightly more then a single PSU would, due to the fans on the second PSU.
Matt: You are correct that one does not sleep. Also, Joe is 100% correct that the smaller fans need to run non-stop.
I just purchased and received a Supermicro SC933T-R760B so I will have a 760w PSU with three 380w modules that I can do another writeup with and illustrate.
This is the marvellous post that I have come over after huge searches. I am really thankful to you for providing this unique information about redundant power supply unit.
Just curious -> can I plug in 2 different PSU’s ??Let’s say one 700 W and one 1200 W device ?? When they share the load, it must function – I think.
Any suggestions ??
BTW: Good article !! THX
can I connect two different APC UPS’s to each power supply ?