post

Day 4 – Wordle

This visual is a combination of the information presented in Day 2 and Day 3.  It is a highly visual, instantly understandable  presentation of which rack is the hottest and which is the coldest.  Using an online site for generating word clouds (wordle.net), I was able to visualize the average rack top temperatures.  The larger the text the hotter the temperature, the smaller the text, the cooler the temperature.

Average Rack Top Temps Wordle

Takeaways

  • Racks 115, 112, 129, and 148 seem to be the hottest
  • Racks 33, 56, 54, 2, 55, and 15 seem to be the coolest

Issues

  • The hottest and coolest racks that I identified above are not exactly quite right.  If you compare to the tables on Day 2, which are numerically sorted, you can see that I missed a few.  Although neat and fun to look at, this visual is prone to error.

Previous Posts
About the Data
Day 1 – From 10,000 Feet
Day 2 – Rack by Rack
Day 3 – Learn to Love Graphs
Today – Wordle
Tomorrow – Bubble Charts

post

Day 3 – Learn to Love Graphs

Top of Rack Temps

Bottom of Rack Temps

Humans are visual creatures. Our ability to identify patterns is a strength that should be used when analyzing data…which means –  graph it!   These simple line graphs showing all 149 rack top and rack bottom temperatures sum up all the data in a comprehensive manner.  These visuals allow us to see every piece of data for the entire time period.  Note that I left off the legend in these graphs to save space.

Takeaways

  • The hottest and coldest racks in the data center can quickly be identified.  Looking at the top temperature graph, it is interesting to note that the coldest rack is significantly cooler than the rest.  I wonder why?
  • Beginning on 12/26 a large portion of the temperatures drop for about a day then spike, then return to their previous position.  Both the top and bottom graphs reflect this trend.  Some racks do not follow this pattern. Perhaps they are in a separate room?
  • The rack bottom temperatures have a much wider temperature disparity, ranging from 50°F – 89°F.  The rack top temperatures range from 58°F – 87°F.  In order to get 89°F at the rack bottom, I suspect there is some mixing going on in that location
  • The graph makes it quick and easy to identify that certain racks are operating above the upper thermal limit of 80.6°F.

Issues

  • Visuals provide a much friendlier way of looking at the data, but they can still be overwhelming.  If the operator wants to quickly identify which are the hottest, and thus most critical locations, he has to inspect the graph and do some mental thinking.  Tomorrow’s visual will leapfrog this step  altogether.

Previous Posts
About the Data
Day 1 – From 10,000 Feet
Day 2 – Rack by Rack
Today – Learn to Love Graphs
Tomorrow – Wordle

post

Day 2 – Rack by Rack

Rack by Rack Temperatures (truncated at 30 sensors for space)

This view shows the min, max, and average values for the top and bottom of each rack.  The actual table has 149 rows, but I cropped it for space purposes.  You’ll quickly realize that while this more detailed information is much more useful than yesterday’s table, it is difficult to extract any actionable information if we can’t sort this data.  Here is the same data table sorted to show the 20 hottest rack and 20 coldest based on their average.

Hottest Racks

Top Coldest

Takeaways

  • 13 racks average above the upper thermal limit of 80.6°F and 1 rack averages below the lower limit of 64.4°F.  The hot and cold spots have been identified.
  • On Day 1 we saw that the maximum rack top and rack bottom temperature were both 88.7°F.  Our question was if they came from the same location.  We now see that they don’t – rack 115 has a max top temp of 88.7°F and rack 20 has a max bottom temp of 88.7°F.

Issues

  • While this information is valuable, presenting it in a table is still overwhelming and difficult to quickly extract the most useful pieces of information.
  • Tables are useful for presenting static information but not very good at showing time dependent patterns.  Min, max, and average provide a good summary for the entire time period, however they cannot reveal any time dependencies.

Previous Posts
About the Data
Day 1 – From 10,000 Feet
Today – Rack by Rack
Tomorrow – Learn to Love Graphs

post

Day 1 – From 10,000 Feet

Data Summary

Let’s begin by taking a very high level view.  The table above calculates the minimum, maximum, and average temperature for the entire data set and display it in a simple table.

Takeaways

  • Maximum temperature exceeds the ASHRAE recommended upper thermal limit of 80.6°F and the minimum temperature exceeds the lower limit of 64.4°F.  It is evident that there are hot spots…and cold spots
  • Average temperatures for rack tops and bottoms are within ASHRAE recommended ranges
  • Top temperatures are hotter than bottom temperatures, as expected, for min value and average value, but not for max value.  How come?

Issues

  • It doesn’t make sense that the top and bottom temperatures would be the same for the maximum value.  Are they coming from the same location?
  • You can’t determine which specific racks are outside of thermal threshold so there is no way of locating the hot spots. Although this information is at a glance simple, useful and high level, a data center operator needs more granular visibility into the statistics for each rack

Are there any other important takeaways you can identify in the data?  Leave your thoughts in the comment section.

Posts in Series
Yesterday – About the Data
Today – From 10,000 Feet
Tomorrow – Rack by Rack

post

Data Confessional

“If you torture the data long enough, it will confess.”

                                                   -Ronald Coase

I love that quote because it gives data a humanistic capability…the capability to obtain useful information. Inside every set of data, no matter how simple or mundane, there is information to be obtained. Data not only contains information that is obvious, but also information that we don’t yet know exists. When looking through raw numbers, charts, and graphs from every perceivable angle and combination, there comes a moment that these small but important pieces of information will emerge. With persistence, an open mind, and a will to engage in a bit of targeted torture, you can make a little bit of data go a long way. Over the next seven days, I will take a set of seemingly simple temperature data and massage it, analyze it, and torture it until it confesses what it knows. Each day I will present the data using a new visual. I will show you how, by analyzing and displaying data in different ways, you gain entirely new actionable information and surprising insights.

About The Data

The data that I will be analyzing is one week of rack inlet air temperatures obtained from an intelligent energy management system in a data center. Temperatures were measured in the cold aisle at the top and bottom of the rack. The data was collected with 149 sensors resulting in 298 monitored points in the data set (149 top temps and 149 bottom temps). The data is comprised of hourly averages for each point for the duration of the week. Below is a sample of the raw data.

Raw Data (Top of Rack Temps)

Raw Data (Bottom of Rack Temps)

Tomorrow will be the first day of analyzing this information in the attempt to gain insightful and actionable information.  Stay tuned.

Tomorrows Post: From 10,000 Feet

 

post

What is an EC Plug Fan?

EC plug fans will do to the data center industry what the ipod did to the music industry.  EC plug fans are, in a word, transformative.  EC plug fans provide  a new means of airflow through a CRAC or CRAH unit more efficiently, more evenly, and with significantly less energy than the currently adopted scroll fans.  In what will be the latest toy for data center operators, the fans will be adopted quickly and enthusiastically. This post describes how these fans work and why they are so much better.

[Read more...]

post

Intelligent Control Harnesses Big Data

What do pilots, electrical grid operators, and data center managers all have in common?  They all control a piece of equipment that cannot fail.  A pilot error, large-scale blackout, or financial data center outage can result in unquantifiable costs.  It is their job to ensure their equipment is operating sufficiently at all times. Given their focus on reliability, it is understandable why, at least in the case of data centers, not much attention has been paid to optimizing the efficiency of the data that is collected for these systems.  This post examines how each operator manages critical equipment to avoid failure today and introduces a new concept of how more intelligent control using big data can improve both safety and efficiency profiles.

[Read more...]

link

Apple just submitted a deed to build a data center in Prineville, OR.  Remember who also just moved in to Prineville?  See a description about the planned Apple facility from Data Center Knowledge here:

http://www.datacenterknowledge.com/archives/2012/02/21/apple-confirms-plans-for-oregon-data-center/

post

How cool should my data center be?

How cool should your data center be?  The answer to that question depends on who you ask.  If you ask the customer they will say: “cold enough so the equipment doesn’t fail.”   If you ask the IT equipment manufacturer they will say: “cold enough not to void the warranty.” If you ask an energy consultant they would say: “shouldn’t be that cold!”  Thanks guys…very helpful.  Luckily, the reason that the American Society of Heating Refrigeration and Air-Conditioning Engineers (ASHRAE) exists is to be able to provide an unbiased answers to questions like this.

However, their answer to the question will be: “well, it depends.”  Fortunately it is ASHRAE’s job to define ‘it depends’ and they have done just that in their 2011 Thermal Guidelines for Data Processing Environments.  This post summarizes those thermal guidelines.

[Read more...]

link

A few months back I wrote about an underground data center in Sweden (see post here).  Clean Technica just published another article highlighting a few more underground data centers:

http://cleantechnica.com/2012/02/13/6-green-data-centers-that-could-survive-a-zombie-apocalypse/