Data Center Power Draw Calculation

Question

I have a difficult time calculating the power draw of our upcoming server (Dell R710) and what that means in terms of how a data center will calculate power costs. Hopefully you can help shine some light on this!

1. I've looked at Dell R710 Power measurements at spec.org which tells me that the server is drawing between 64 (idle) and 236W (full load). Our server's specs are slightly below the measured server, so I expect the draw to be even less.

2. I understand that Amp are calculated by Watts / Voltage. In the United States, the outlet voltage is 120 V. That means, the server is drawing between 0.5 and 1.9 Amp depending on load.

3. Most data centers I've looked at say they're including 1.5 Amp with a 2U space. Is this based on average draw? When is it typically measured? What happens when the server draws more power?

4. The smallest R710 PSU is a "570W" PSU - but it looks like it will never, ever, require 570W. How does the PSU relate to the power draw?

This is kind of confusing, please help me understand how this typically works a bit better. I have a feeling I'll run into power overcharges otherwise!

Edit: Even more confusing, Dell's Energy Advisor calculates 3.1 amp for my server (Dual E5620, 24 GB RAM, 4 SAS HDDs). I'm getting lost.

Solution

On the Dell R710 (and many other makes/models) you can monitor the power usage yourself with this command:

# ipmitool sdr list | grep Watts
System Level     | 84 Watts          | ok

That's the linux version, but there are Windows equivalents. Graphing that in your favorite tool is left as an exercise for the reader. It should be noted that this gives the power draw into the motherboard. Power supplies are not ever 100% efficient, so add about 15% to that number to get the input into the power supply. Or connect it to a watt meter and measure the efficiency yourself. PSUs are most efficient in the middle of their stated range, somewhere about 50-60% of the rated capacity.

If you are concerned about power usage you might consider using an L-series processor.

What happens when you draw more? That depends on the provider. You'll likely just get a warning (if they even notice at all.) And that's also the scary part. What if everyone draws just over and the circuit breaker trips? How closely do they monitor those circuits? Is it active monitoring or passive monitoring (is there a meter on the circuit or do the building engineers do spot checks with a clamp on meter?) If there is a meter is it per power port or per circuit?

Overall, it's just best to monitor the draw yourself.

How do you know before you order the server? Well, that's a guessing game. Unless you're really cranking on the HW you won't get near the peak.

Comments

[Admin]

Are you sure your datacenter is providing 120V power? Many are providing 208V now, which would be much preferred if your equipment can handle it (which is quite likely).

[Admin]

Not completely sure, I will check first thing in morning. Thanks for the tip!

[Admin]

I can tell you our R710's forced us to upgrade to the larger PSU's when we added a second proc and another 24GB of RAM to make it an ESX host.

[Admin]

Tom, what are the specs and the power draw on your server? Would be very interested to find out.

[Admin]

We've been doing a lot of shopping around for rack space in the past month or so, and we've always seen them advertise their power capacity in kVA, not just amps. Maybe that's a southern hemisphere thing?

[AX1]

Thanks for the info! Do you know how data centers usually deal with overages (e.g. drawing 2.5A instead of 1.5 during a peak period, and then less than 1.5A for the rest of the day)? Shut off the server?

[Luc H]

>AX1 very old post, but still replying as i came upon this question. Most Datacenters use per-kwh pricing and just have you pay more if you go over their allocated kwh in a month. Discuss this with your datacenter as it might differ for yours.