Load Study for Data Center

[Senselessticker]

I'm working on a load study for a "data center".

Basically, the task is to determine if an existing power distribution system is capable of handling some additions to the server racks.

I know from years of experience how difficult it is to calculate electrical loads for computer/server type rack mounted equipment based on nameplate load info. The client is pushing for an entire upgrade for the UPS, generator, etc.. that feeds the server room (about 1k sq ft rack room). Does anyone know of any good standards to follow for designing/justifying power available per sq. ft. for such type of rooms?

Quick internet searches today show that such rooms are designed anywhere from 50 W/ft^2 to 400W/ft^2. My "gut" (a.k.a. experience) tells me that the existing power system is perfectly capable of carrying addition loads in racks. In my experience... computer/server/network type equipment generally operates at about 10% of its nameplate input power. But I need a standard to follow to show the client this... The client is not satisfied with simply field "measuring" the existing demand load, then working from there...

Any advice? Or any IEEE standards out ther that can help me out justifying W/ft^2 for server type rooms?

Thanks!

If it were "my" data center, I would not even hesitate to say: "the existing power systems is just fine.... plug it up!!!". But I need a standard to follow to show the client this...

 

[davidbeach]

If the client wants it, why are you holding them back?

 

[alehman]

The power density range you mentioned is reasonable as an average. There is no "standard". It depends entirely on the type of equipment and density. You have to examine each case on its own. In response to davidbeach, I think if you can substantiate that an upgrade is not needed, there is a professional obligation to inform the client accordingly. Then if they say shut up and do it, then do it. (I am facing a very similar situation with a project now).

 

[unclebob]

As a service guy, I would put a recorder for a week to measure exactly what is the current load. Knowing your transformer and electrical equipment maximum load and your power consumption profile, you know how much you can add.

You can, by the way, measure your harmonics, power factor...

 

[Laplacian]

Is the power for the load center metered seperately by the utility? If so, they may have peak demand on the bills and you can go back and trend historical peak demand for each month. Even if it's not seperate, it may help to determine the load factor.

 

[Senselessticker]

The power coming into the load center is filtered and distributed through a Power Distribution Unit (PDU). The PDU monitors and records load demand. I certainly plan to use this information as part of my report and calculation. The PDU is running about 10% capacity right now. Peak Demand I think was about 12% when I ran through the menu on the PDU.

The problem is that I'm getting the feeling that the client wants me to use nameplate info for individual server equipment as a basis for determining additional load on the system. For instance, some of the server equipment power supplies are 8kW. My guts tells me these servers will probably only draw, say 500W or so. It would be very helpful if I had a standard to follow which could help me determine demand and diversity factors to show this on paper.

Dealings with this particular client in the past has proven to me the client lacks a great deal of common sense and fundamental understanding of electrical power distribution. Without a real standard to follow, I can certainly tweak the calc to lean either way. I just have to get the client to buy into the method used.

The idea of doubling/tripling the power capacity to this one rack room could have the following consequences: new UPS required, new generator required, new HVAC system for cooling, New/larger electrical room to house the additional equipment and UPS/generator, you can imagine the architectural nightmare, construction cost, etc.. etc... could easily turn into millions of $$$$) just to power up a number of network servers, etc... The client could almost build a new facility for what it will cost up upgrade the fairly new (a few years) existing one.

I know that ASHRAE has developed a technical committee to look into better design methods for data centers (I've personally seen cooling systems designed and installed which carry about 10-20-30?? times the available cooling capacity than what is really needed for a data center type room/facility). Same goes for a bloated power system... And all a result of poor design fundementals and lack of good "engineering sense". Anyway, Does anyone know if ASHREA TC 9.9 ever got anywhere?

Thanks again..

 

[BJC]

senselessticker
The one I just fisished is 225 VA/ft-sq. The next generation is going to be 400.
One thing to keep in mind is that ALL of that 225 VA turns into HEAT. None of that power can get out that Fiber Optic line.
Without the CRAC units and chilled water system the place will last about 10 minutes then servers start cooking. Circ-pumps are on UPS and there is a large chilled water storage tank. The chillers are all on D/G backup. Reminds me of core melt situation.
I would find out what reliabiliy the clinet wants and is willing to pay for. N+1 or even N+2 won't get it for most data centers. 2N+ is they way I see them.
You need to be able to service EVERY component in the place with out going down or working anything hot. In addition you will want to be able to load test every thing off line.
If your client is willing to spend the money do it right. IF not build it as cheap as you can and tell them to expect outages. Building one halfway is like building half a bridge- not much good.
The Googles, Intuits, Yahoos, MS etc. show up with dumptrucks full of money. Of the choice between Good, Cheap and Fast, cheap ain't a word they know.