Home Page

  Recent Posts

  Recent Comments

  Archives

  Categories

  Contact Us

sustainableIT (Pty) Ltd

Head Office
Tel: +27 (0)21 686 6085
Website:www.sustainableIT.co.za Website:www.thecarbonreport.com Email: Head Office


  Subscribe to our newsletter

  1. (required)
  2. (valid email required)
 

cforms contact form by delicious:days

 

Getting more out of PUE in the datacentre

Datacentre efficiency is at the forefront of most green IT initiatives and many organisations are embarking on programmes to green the data centre.  As with most sustainable initiatives, it is extremely important to baseline where you are before making changes so that you can measure the benefit and business savings.

Datacentre Efficiency

The most common measurement banded about is Power Usage Effectiveness or PUE, developed by the Green Grid which includes leading IT companies from all over the planet.  This is measured by calculating the ratio of total energy consumption (servers + cooling) to ‘useful’ energy consumption (servers only).

PUE = Total Facility Power/IT Equipment Power

A typical enterprise data center is thought to have a PUE of 2.0 or greater. This means that for every watt of IT power, an additional watt is used to cool and distribute power to the IT equipment.  The first problem with this is that nobody knows what a good PUE is and the Green Grid would have us believe that this will come over time as more and more datacenters are measured and best practice is defined.

The more fundamental problem with PUE however, is that it takes no consideration of useful work, which by definition is what efficiency is all about.  Unless you have some sort of measure of the useful work being performed by a datacenter, the measure is actually meaningless.

The best way to elaborate on this is by way of example:

Scenario 1

 We have a datacenter that is world class, the latest technology around power and cooling, blade servers, shared storage and a virtualized environment.  For illustrative purposes, let’s assume that the PUE of this datacenter is 1.3, in other words:

Total Facility Power (1.3) / IT Equipment Power (1) = 1.3

At face value this looks good but now let us assume that the useful work being performed by the servers is actually only 10% of the processing capacity, an inefficiency of 90%.  Adding this to the equation we get an efficiency rating as follows:

Total Facility Power (1.3) / IT Equipment Power (1) * Effective Inefficiency (0.9) = 1.17

Scenario 2

 We have a datacenter that has not been optimised with new technology but the devices are used optimally, in other words, the IT department is squeezing every ounce out of the infrastructure that it can. For illustrative purposes, let’s assume that the PUE of this datacenter is 2.5, in other words:

Total Facility Power (2.5) / IT Equipment Power (1) = 2.5

At face value this looks bad but now let us assume that the useful work being performed by the servers is actually 80% of the processing capacity, an inefficiency of only 20%.  Adding this to the equation we get an efficiency rating as follows:

Total Facility Power (2.5) / IT Equipment Power (1) * Effective Inefficiency (0.2) = 0.5

The lower the PUE the better and with this methodology, let’s call it Power Usage Effectiveness and Inefficiency (PUEI), we are bringing into the equation useful load.  At face value inefficient datacenters may be doing more useful work and hence are actually more efficient.  Although these examples are artificial, the reality is that I would rather be running the datacenter in scenario 2 as it is delivering more and in reality is more efficient relative to useful work.

As with most hypotheses there are problems with this, the most fundamental being measuring useful work.  In our opinion, Green Grid members would be best served by concentrating their efforts in this space than anywhere else.  Unless we can measure useful work, it is hopeless talking about datacenter efficiency.

There is light at the end of the tunnel however.  One member of the Green Grid, 1E, is soon to release a product called NightWatchman for Server which will effectively measure useful work performed by Wintel and Linux servers.  Not only will this give us the ability to assess useful work but also provide the information required to make decisions around decommissioning, virtualization and consolidation.  Until these types of toolsets become more readily available, datacenter efficiency in the form of PUE is flawed.

Tags: , , , ,

Leave a Reply

 

 

Home  |  About Us  |  Product Range  |  Savings Calculator  |  Resources  |   Contact Us

sustainableIT | Energy saving insight around your Information Technology

Afrigator