Take the 2-minute tour ×
Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

I've struggled with sizing my power sources and associated PDU's in my server racks. Somehow I feel like I'm way oversizing things based on nameplate requirements on power supplies.

My simplistic approach is to look at the nameplate draw on my power supplies, add them up and size my feed for that. Doing that with a rack that has 10 servers, 1 storage array, and 4 Cisco Switches leaves me with 6 - 20 amp circuits? How do I arrive at this figure? Well, I look at absolute worst case which in my mind is half of my power supplies have failed and I'm starting up all my equipment at the same time on 3 circuits connected to the half of the power supplies remaining. Because I have to assume all load will be concentrated on half of my outlets that effectively doubles my requirements.

Does anyone have a better methodology than this? My customers freak a little bit when I tell them I need 80 or 90 amps for 8 or 9 servers and some switches. It does sound a little over the top. I know one option might be to configure the high draw equipment (like servers and arrays) to not power on when power is restored. I don't know if I have an option to delay power on either. That would be nice. Do switched PDU's have a feature like this where it will stage the outlet power ons. That seems like it would be effective.

Clarification

I know I can use the calculators to determine steady state power draws which are much lower than the startup power draw. My ultimate concern is startup power draw for all components at once. From there I effectively have to double the amount of capacity I deliver to the cabinet to account for starting up on only half of the available capacity because half of my power supplies are failed.

Based on some research it appears switched outlets that can stage themselves might be the answer. Others agree or have better solutions?

share|improve this question
    
Some servers can be set to have a randomly-generated power startup delay. –  ewwhite Jun 8 '11 at 14:42

4 Answers 4

I use the HP Power Advisor and the APC UPS Selector tool to provide a good estimate of power requirements. The APC tool has a great database of common telco/server equipment, and provides a way to build in some headroom for expansion.

share|improve this answer

The way I tackle this problem is to ignore the nameplate information and actually measure the current draw of each piece of equipment. There a are a number of ways to do this but the simplest solution for me is to use a power lead which has been modified so that one of the conductors forms an accessible loop of wire outside the cable. A clamp meter is then used on that wire. Measurements are taken both at startup and while running, under both normal and heavy loads. With this information I can then work out realistic power requirements. If you're unable/unwilling to do this yourself talk to an electrician about it.

share|improve this answer

The nameplate draw is going to be maximum configuration worst-case draw. I'll second the power advisors that ewwhite suggested (I use the APC one when building out racks myself), and using a power meter to figure out your actual configured power use, if you have the equipment already (or if you can get eval equipment to test with).

Also, the storage array and switches may vary widely in power use. I had enterprise storage arrays that warranted 2-3 dedicated 20A circuits each; if your array has multiple controllers and dozens of drives, it's different than, say, a 1u 4-drive storage array. Often the vendor will have recommended practices, and the vendor SE should know how accurate the published specs are and what people are actually doing as far as power distribution. Ideal is not always practical.

I've used things like Kill-a-watt (which plugs into the wall and your server plugs into it) to evaluate single power supply load, or dual with a split power cable. Some of the metered/managed PDUs will report on current draw (pun not intended to be funny) so you can figure it out without unplugging/rebooting your server. But if you have dual power supplies into separate PDUs, which is sensible, you may not get the whole picture.

share|improve this answer

I don't know what your voltage is, but 90 A would only sound sensible with something like telco-rack 48 V equipment.

Rather than looking at the power supply rating, you should be looking into the data sheet for the specific system for a better estimate on power consumption. The supplies already come over-engineered, so just because a system runs with 2x 1200 Watts PSUs does not mean that it will suck down 2,400 Watts at any point in time - especially since those would be running in a redundant configuration which are meant to provide enough power for every situation even in the case of failure of one of the PSUs.

APCs switched rack PDUs do support a "staggered" powerup - you can set a powerup delay on each outlet. Other manufacturer's gear might do that for you as well, but I only ever worked with APC stuff. Apart from that, most server systems and disk arrays (or rather their controllers) also would not power up all drives at once - this helps keeping current spikes down.

share|improve this answer
    
1kw (max) per 1U server times 42U of servers is 42kw. At 110V, that's ~380A (if my math is right - but it does seem high) –  warren Jun 8 '11 at 18:16

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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