Transcript

Introduction

Hello, and thank you for joining today’s Ask a Pro Webinar. My name is Joe Fackler, a DevOps guy here at PMI. Today we’ll be talking about a white paper on Data Center Power 4.

Some of you may have joined our Data Center Powers 1, 2, and 3. Don’t worry, they’re not a requirement, although I do encourage you to go take a look at them if you have your utility supporting data centers, or if you’re just, in general, curious about some of the utility perspective and concerns regarding data center power.

High Speed Protective Devices in Data Centers

In Data Center Power 3 we did mention that there are very high speed protective devices installed at modern data centers. They are incredibly fast operating on the microsecond scale, and they kick over to the backup device or to surge devices. Sometimes these are even hot swap live generators now that they have ready to be on site.

I think it was Data Center Power 2 we talked about how one instance in Northern Virginia saw a one and a half megawatt demand just drop immediately, and the utility load calculations are suddenly very different.

A Measurement Blind Spot

Specifically, we’re looking at a measurement blind spot with mysterious protective device or breaker operations inside of a data center that may get missed by some of the older traditional power quality monitoring. A high tier data center customer calls in, there’s a main breaker tripped or sub breaker trip or something kicked over to a backup power.

You go out, you place a recorder, you come back after a while, you look at the recording data and you see everything looks pretty normal. There’s no sag, no swells, no sustained distortion. And these trips can be described as nuisance or mysterious, but we do know that the protective equipment did operate the way it was designed.

RMS Averages and Older Monitoring Equipment

Some of the older power quality monitoring equipment is RMS averages. And this is not just on the utility side. This is not just something that utilities place out in the field that records power. This is also some of the older equipment that data centers have.

They have equipment set up in their PDUs, now their rack power distribution that’s SNMP polling, where some central system is querying every so often, “Hey, what is the power light coming in? What’s the power light coming in?” So, if you’re looking at some of these devices that take, they may sample at decent rates, but they’re recording an RMS one second average or similar kind of average.

Here in this first graph, we’ve got a RMS average during the, wherein the system is actually experiencing a one megahertz transient. You can see it barely moves the needle on the RMS average. So, if you were to pull this back and look at the data, or if it was pulled off a recorder built in, SNMP type recorder built into the data center rack itself, you wouldn’t see that, or they might not see it on there.

And yet they’ve got protective equipment firing, they’ve got backup power getting kicked on, but the RMS average has missed it entirely. So this is where sampling rate becomes very important. Not just sampling rate, but how you can display that sampling rate.

Sampling at 256 Samples Per Cycle

Again, remember, this is the same transient. This is zoomed in on the same graph. If you were sampling at 256 samples per cycle, which is a fairly common sampling rate for power quality equipment, you can see here, these little red dots are the samples being taken.

Now, you’ll notice we caught some of the transient. This is a one megahertz transient. We caught some of it. And when you go to graph that, you’re not going to catch the full amplitude of this disturbance, but you would catch something. So all the power quality equipment, Revolution Seekers, Voltz, are all capable of recording these samples per cycle. And so you would catch that in the waveform capture. You’d see some of it, maybe not all of it.

Now, I intentionally, creating this graph, put that transient peak at exactly the worst possible time, right dead between the samples. But that could happen. The more expensive the equipment on the other end, the more likely you are to get the worst case thing to try to catch.

Full Transient Capture

Moving on down to a full transient capture. Recorders like our Revolution are capable of detecting a one megahertz transient, and not just giving you the 256 sample view of that waveform, but actually catching that entire transient waveform.

So here, this final graph is saying, “Hey, here’s a chart where you actually were sampling at a fast enough rate to catch the entire transient.” Those of you familiar with the Nyquist know that you have to be sampling at least twice as fast as the sample you want to catch.

You can see here, again, same transient. I’ve drawn a little red line right at where that protective equipment is going to fire, where you’re gonna trip that threshold, you’re gonna kick the data center over to utility power to protect the sensitive equipment on the other side.

And our one second RMS average would catch none of it. Our 256 samples per cycle would catch some of it, but not necessarily the peak. You might get some clipping if it was just at the wrong time. The right time, you might get the exact peak dead on, but you can’t guarantee that.

Understanding the Full Picture

So this is a zoom back out here. We’re not just looking at the graphs, but the graphs are useful to give you a visual understanding of what we’re discussing. We’re talking about very fast moving transients, because in Data Center Power 3, we talked about some of the high speed inverters and devices that are present in data centers can kick off these very high frequency harmonics, and indeed transients, which are hard to catch.

And depending on what system you’re using to capture them, you might miss them entirely, you might miss the full magnitude, or you can capture the entire transient from beginning to end to match up between what the utility is seeing, what the data center is seeing, and even between what the data center protective devices are seeing versus the data center power recording equipment. So you need high speed transient capture equipment to properly catch something like this.

Trip Characteristics and Protective Equipment

We’ve got some more trip characteristics defined here in the paper, instantaneous versus delayed operation. It will be important if you are a utility sent out to deal with equipment like this, that you understand and find out what the protective equipment the data center is actually using and how it responds, what it does, how quickly that it is sampling on the protective side, because you need to be tailoring what you are monitoring and capturing to the receiving side equipment, what it is, where it’s designed to trigger.

Closing

Thank you for your time today. We looked at some graphs, and if you come up with a question after we end the webinar, you can always give us a call at 800-296-4120 or by emailing support@powermonitors.com. Thank you and have a good afternoon.