Improving Data Center Airflow Managment Using Analytics

This data center has poor airflow management (lots of mixing issues)

data center with lots of mixingThis data center has efficient airflow management (little to no mixing issues)

data center with no mixing

Big bubbles in red = mixing

Big bubbles in green = no mixing

Feel free to stop reading at this point, because this is really all you need to know about mitigating poor airflow management in your facility.

But, if you are curious as to how this analysis was achieved, and how analytics like this can be used to turn mountains of data into actionable information, read on.

The following bubble chart displays the server inlet temperatures collected every hour at the top and bottom of 30 racks, over 7 days.  The chart contains 10,260 individual data points, accordingly.

Chart Description

The Y-Axis on this chart indicates the difference between the top and bottom temperature points.  I to refer to this as the delta-T.  Mathematically it is: top temperature – bottom temperature

The X-Axis lists each of the 30 racks (labeled 1 through 30).

The bubbles indicate the delta-T at each rack.  There are multiple bubbles for each rack because the delta-T changes over the course of the 7-day measurement period.

The size of the bubble indicates the length of time each rack maintained that particular delta-T.  Larger bubbles indicate more time and smaller bubbles indicate less.

The color of the bubbles are displayed differently side by side for visual acuity only.

data center bubble graph

How do I read this chart?

Start with Rack 1 which displays the dark blue bubbles to the far left.  During the 7-day date range, Rack 1 had a delta-T (top temperature – bottom temperature) that fluctuated between 4°F and 6°F.  The majority of the time the delta-T for Rack 1 was 4°F, indicated by the larger sized bubble.

Now look at Rack 9.  The delta-T for Rack 9 fluctuated between 4°F and 14°F, but it spent most of the time at 12°F.

Rack 12 spent the entire time (i.e. all 7 days) time at -2°F.

What does this data tell us?

In most data centers, we expect the temperature at the top of the rack to be hotter than the temperature at the bottom of the rack. This is simply because hot air rises.  This will even be the case in an overhead air distribution environment (but perhaps not as much so as an under floor distribution).  However, since we are comparing the data to the rest of the temperatures in the facility, a marker of good analytics, this analysis will provide the same insights in any type of environment.

This particular facility had under floor air distribution. We would expect that the delta-T be somewhere between 0°F – 8°F.  Said another way, we wouldn’t expect the top of the rack to be more than 8°F hotter than the bottom of the rack.  And certainly, we wouldn’t expect the bottom to be hotter than the top, which would be indicated by negative values.

The green region indicates the expected range of delta-Ts.  The red region indicates delta-Ts outside of this expected range.

data center bubble graph

It’s clear that over the course of the 7-day period there were some delta-Ts significantly outside of the expected range.

The beauty of this analysis is that not only are those abberations clear, it also becomes simple to identify the specific culprits by highlighting which racks have big bubbles in the red region.

data center bubble graph

The racks highlighted in gray on the X-Axis have delta-Ts that lie outside our expected range.

Let’s start taking action on the data

Why are the delta-Ts for these racks outside our expected range?

There could be a number of reasons, however it’s usually an indication of mixing.  Some reasons might include:

  • No blanking panels
  • To many or too little perforated tiles
  • Too much or too little airflow
  • IT equipment that is installed backwards (and blowing hot air on the bottom sensor)
  • Racks that do not conform to a cold aisle/hot aisle layout

Analytics may not always be able to tell you why something is occurring, but good analytics can indicate where your problems lie so you know where to focus your attention.  Visually inspecting the racks highlighted in gray will probably reveal the root cause of the mixing issue.


What does a graph look like indicating minimal mixing?

Re-running the analysis after implementing some simple airflow improvements for the culprit locations will look more like this:

data center bubble graph

But I want my airflow to be perfect

Then see if you can achieve this:

data center bubble graph

With the right rack layout, containment strategy, floor/rack/cable seals, blanking panel management, perforated tile layout, and cooling control, this is achievable.

And even better, you can see it, track your progress, and prove it to your management or colo clients.

I challenge you to get there.

More Interesting Information

3 thoughts on “Improving Data Center Airflow Managment Using Analytics

  1. Pingback: Fan Law — Data Center Huddle

  2. Good practice and article! I help many clients analyze the inlet temperature and manage the energy using by this way.

Leave a Reply

Your email address will not be published. Required fields are marked *