datacenter: 80% confusion

Status
Not open for further replies.
S

ss

Member
Hi,
I've got a question about circuit loads. We are planning on installing a rack of 12 render nodes (like servers). Hewlett-Packard says that the real world load that each of these servers draws is 1.9 Amps at 208 VAC. We were planning on using a 208 VAC 30 Amp circuit to power all of these servers. My impression of circuit loading was that you could go up to 80% of what the circuit is rated at. Since each of the servers is 1.9 Amps and there are 12 of them, this would put us at 22.8 Amps or 76% of the rated circuit load. The building engineer has taken issue with this purposed setup saying that 76% of the load too close to the 80% specified by the NEC.
I was always under the impression that you could load up to 80%. He seems to be suggesting that 80% is a drop dead max and that you should plan for a lower percentage of circuit use.

What the right answer? What Amperage percentage should be shooting for?
 
Re: datacenter: 80% confusion

Very doubtful it would be code compliant to feed these units from a 30 amp breaker.

Are they cord and plug connected?

What size plug is on the unit?

If it is hardwired what does the UL required labeling say?

You have to forget what Hewlett-Packard says about the real world load. The branch circuits must comply with the UL labeling.
 
Re: datacenter: 80% confusion

By the way I do agree with you that the NEC would allow the circuit to be loaded to 80% but that will be 80% of the label or nameplate not what Hewlett-Packard says is the real world load.

Now the next thing that comes to mind would be does the customer want these units on multi outlet circuits?

Most units in a Data center are on individual branch circuits and sometimes have redundant circuits feeding them.
 
Re: datacenter: 80% confusion

Iwire,
The proposed system would be setup as follows. There would be 12 computers (servers) each with a single power supply. The power connector on each server is an IEC 320 C13 connector (a standard PC power inlet connector) and each server comes with a detachable 10 Amp rated IEC 320 C13 Male to Female cable for 208 V operation (they come with an IEC 320 C13 Female to Edison/U-ground 10 Amp cable for 120 Volt operation). They would then be plugged into an APC Rack PDU rated for 30 Amp 208 Volt operation. The APC PDU has IEC 320 C13 output connectors and a NEMA L6-30P input connector. This input connector would be plugged into an overhead raceway mounted L6-30 receptacle that is fed by a 30 Amp breaker in the panel.

The type of application that these servers are running is not one that requires ?high availability? so the customer does not need/want any sort of redundant power configuration (additionally, these server are not being purchased with redundant power supplies).

I understand and agree with what you are saying about UL nameplate rating vs HP?s ?real world? rating. The thing that confuses the issue is that HP?s Power Best Practices documentation has the following paragaph:

"The amount of power a facility will need can be calculated using one of three methods:

-Using nameplate ratings ? This method worked in yesterday?s environment, but it is a costly method resulting in wasted infrastructure dollars.

-Using actual power measurements ? This is the most accurate approach, but numbers are difficult to generate and collect when in a planning mode.

-Using ProLiant Power Calculators ? HP recommends this as the best practice for advanced planning since it produces more realistic numbers. "

In looking at this, HP?s assessment of this rings true for us. The actual power measurements that we have done and the HP Power Calculator are in somewhat of an agreement but nameplate ratings are well above. I think part of the reason this happens is because the nameplate rating doesn?t change based on what options are installed in the server. A server with all the minimum options has the same nameplate rating as a server with all options installed in it (even though those options will draw more power).

I would be interested in further discussion on this nameplate issue as well as on my original question. Thanks very much for your help so far!
 
Re: datacenter: 80% confusion

Unfortunately you must use the nameplate data for your calculations. Just like you can put a 25W lamp in a fixture rated a maximum of 150W and the real world draw will be 300W, because the nameplate maximum says 150 that circuit will be able to draw 1800W. I think the engineer has you on this one.
 
Re: datacenter: 80% confusion

ss sir,
just my .02 cents----- you are looking for trouble. oh, it will work, but eventually, your gonna have problems. i have been a certified infrared thermographer since 1989, and have seen this type installation. computer loads are fairly constant and when you force 80 per cent of a breaker's rated load over a long period of time, it breaks down. then consider voltage drop to the computer --- current draw is a major factor in voltage drop. is a computer something you want to apply low voltage to? as the breaker begins to fail, the contact points "heat up" and what does this cause--"more voltage drop"! consider the cost of these computers, also consider the down time when the breaker begins to trip on a power bump. most computers perform an important function --- is this system important enough to provide a secure and proper installation? and when you start having problems --- "usually" there is not a spare 208 volt outlet close by to use temporarily until the additional power is installed. i would go with three dedicated 208 volt circuits terminated in duplex receptacles --- four computers on each circuit--- and guess what-- in a pinch you could double up eight computers one one circuit until the problem is corrected.
 
Re: datacenter: 80% confusion

Am I missing something?
The computers will be plugged into a UPS. The UPS has an L6-30 plug, therefore the branch circuit can be no larger than 30A.

Now if you want to argue that the UPS will be overloaded that is a separate argument and one which I think is worth pursuing.
 
Re: datacenter: 80% confusion

If these are computer loads, constantly running, you should use size your load at 125%, which means you cannot load your circuit past 80%.

In response to these manufacturer comments:

"The amount of power a facility will need can be calculated using one of three methods:

-Using nameplate ratings ? This method worked in yesterday?s environment, but it is a costly method resulting in wasted infrastructure dollars.

-Using actual power measurements ? This is the most accurate approach, but numbers are difficult to generate and collect when in a planning mode.

-Using ProLiant Power Calculators ? HP recommends this as the best practice for advanced planning since it produces more realistic numbers. "


These are design issues and should not be taken into account when sizing breakers or branch circuit conductors. They can be taken into account to prevent greatly oversizing your ups, generator system, pdu's or distribution transformers and associated equipment. Eventhough what is oversized today, is undersized 6 years from now.

Jim, I don't think we are talking about point of use ups "shoe box" systems. If we are then I'm not gonna reply anymore cuz they are embarrassing.

[ February 18, 2005, 10:16 AM: Message edited by: bonding jumper ]
 
Re: datacenter: 80% confusion

My error in mentioning a UPS.

I think ss is saying all of the computers are plugged into a single 30A APC PDU.
 
Re: datacenter: 80% confusion

Hi all thanks for the responses - they have been very helpful and informative!

Jim, just to clarify - I was not talking about a UPS of any sort. The device I was speaking of was a rack power strip with no UPS capabilities. The panels that feed this room are fed by a facility wide UPS and generator system.

In terms of the other comments - I now understand that nameplate data should be used for sizing breakers. I'm still a little fuzzy on the 80% thing though. It seems like the general consensus is that, while it is possible to drive the breaker right at 80%, its probably not the best idea in the world - especially over a long period of time. So my question is this: What percentage of a circuit Amperage do you recommend using? In other words, for planning on this project and other projects, what sort of load target should I be trying to achieve? If 75% to 80% is too high, would it be something more like 70% or 60%? Nothing has been purchased yet so we have the ability to do this right ? plus we have many future projects?

Also, if anyone could explain how sizing my load 125% plays into all this. I see that number has come up a few times on this message board. I don't really understand when to apply that as apposed to 80% de-rating a circuit. If anyone wants to share any examples of how to properly plan this, I would love to see them.

Again, thanks for all your comments. It?s great to have so many experts all in one place.

-ss
 
Re: datacenter: 80% confusion

Loading a circuit to 80% of its breaker handle with continuous loads, is that same as saying to apply a 125% factor to the continuous loads when sizing the breaker handle value.

Most circuit breakers will operate with 100% of its rated current (handle trip value) continuously for about 1000 seconds or so, before it trips. With that in mind, your computers are somewhat continuous loads, and should be applied to a circuit with the 125% factor, or pick a breaker that only loads to 80% of its handle, same answer.

I think considering calculating/loading with nameplate data is the way to go, since it is very conservative. I wouldn't have a problem approaching 80% of the circuits ampacity due to the conservative nameplate information.

I often talk with UPS guys whom will tell the first guy that the unit can handle an outage for 30 minutes. That guy he told will tell someone else 15 minutes to cover his butt. Then that guy will tell me 7.5 minutes so as to be sure he doesn't get in trouble during an outage. It is easy to over design and provide misc. future capacity, and difficult to design a system with proper research and limit over design $$$.
 
Re: datacenter: 80% confusion

Originally posted by ron:
I often talk with UPS guys whom will tell the first guy that the unit can handle an outage for 30 minutes. That guy he told will tell someone else 15 minutes to cover his butt. Then that guy will tell me 7.5 minutes so as to be sure he doesn't get in trouble during an outage. It is easy to over design and provide misc. future capacity, and difficult to design a system with proper research and limit over design $$$.
Click to expand...
And then the outage comes and the UPS operates for an hour because it was 50% loaded :) Nice job explaining Ron.
 
Re: datacenter: 80% confusion

ss, just install a larger (or multiple) PDU and branch circuit and be done with it. Are we talking about THAT much more money? I think you are analyzing this way too much trying to figure out what you can get away with. My motto is if you over design it now it won't come back to bite you in the a** later.


-Hal
 
Re: datacenter: 80% confusion

hbiss,
No I think you are misunderstanding the point of my questioning. For this project, cost isn't really a problem - I'm not trying to get away with anything. Adding more branch circuits wouldn't break the bank. For this project that is probably what we will do.

I guess what I am looking for is a good rule to have for future projects that may have more or less equipment. I?m fine with over-designing things, but we need a baseline of where to design things so that we can then over-design based on that baseline. (that?s kind of a tongue-twister - I hope that makes sense)

Thanks!
 
Status
Not open for further replies.
Top