David Cappuccio

A member of the Gartner Blog Network

David J. Cappuccio
Research VP
6 years at Gartner
41 years IT industry

David J. Cappuccio is a managing vice president and chief of research for the Infrastructure teams with Gartner, responsible for research in data center futures, servers, power/cooling, green IT, enterprise management and IT operations. Read Full Bio

Coverage Areas:

Rack Unit Effectiveness–A Useable Data Center Metric

by Dave Cappuccio  |  November 9, 2012  |  1 Comment

Data Centers have gotten a bad rap of late, with both the press and senior executives putting pressure on them to improve overall efficiency and reduce operating costs (yet again).  The focus for the last few years has been around energy efficiency and the PUE (Power Usage Effectiveness) metric developed by The Green Grid.  PUE is a great metric, and when used wisely it can help organizations easily improve energy efficiency oftentimes by 20% and more.  Unfortunately for Data Center managers, the efficiency gains attained with a PUE focus were almost all on the Facilities side of the equation, and while IT and the Data Center may have gained some benefits (e.g. improved cooling), the lion share of the operational savings was applied to the Facilities budget (unless of course you’re one of those rare companies that gives IT it’s own power budget).

But realistically the headlong rush to better PUE’s has done little to improve data center efficiency.  Having a facility with a great PUE is one thing, but if my data center is highly underutilized, or if the resources are poorly managed, IT has not solved the real problem of trying to get the most out of the resources we already have.  The other problem with PUE is that as Data Center managers strive for more energy efficient IT equipment, they could inadvertently degrade that wonderful PUE the Facilities team reported last year.  Take a hypothetical data center with an average PUE of 1.5.  If IT decides it’s time to do a technology refresh on some servers, and bring in the current generation as replacements, the overall performance and productivity of applications will increase, but because of the energy efficiency improvements vendors have made, the overall power draw for IT could very easily go down.  When that happens the ration of total building power to IT load gets worse – negatively impacting PUE.  So a great decision by IT could easily create a bad impression of Facilities, unless everyone understands the overall value of what was done.

So given that rather long preamble, I’ve been thinking about taking the same concept of PUE (optimum vs. actual usage) and applying it to the Data Center proper in order to create a resource efficiency metric.  The problem with creating a metric like this is that all Data Centers are not created equal – and don’t have the same type of equipment or configurations.  So given that caveat, I’d like to propose creating a metric around the most common resource available in most Data Centers – the Rack Unit, or RU.  A standard rack today has 42U, others have 48U, 50U and more, but the single RU itself is something we can track.

So here is the basic idea – and I’ll be writing more on the RUE and RUiE metrics in Gartner’s published research.  Let’s assume the following just for illustrative purposes:

300 Rack maximum capacity (approximately 9,000 feet of floor space).
Standard 42U size
180 racks are currently installed, and average 65% utilization.

The maximum RU count at capacity is 12,600 (300*42)
The Installed RU count is 7,560 (180*42)
The utilized RU count is 4,914 (7,560*65%)

Using the same construct as PUE, we take the maximum and divide by the actual (12,600 / 4,914) and come up with a ratio of 2.56 (where capacity would be 1.0)

The Data Center RUE is now 2.56 and can be track fairly easily to monitor both growth and efficiency.

Using the reciprocal (1/RUE) yields your utilization;  RUiE =  1 / 2.56 = 39%.

Now the big flaw here is an obvious one – nobody wants to get to perfection.  An RUE of 1.0 would indicate you were completely out of room – and that’s not a metric I’d want to attain.  However, using this same idea you could modify the maximum capacity to an optimal capacity. Lets assume no rack should exceed 90% capacity as a target.  The results would look like this:

The maximum RU count at capacity is 11,340 (300*42*.90)
The Installed RU count is 7,560 (180*42)
The utilized RU count is 4,914 (7,560*65%)

RUE now becomes 2.31 (11,340 / 4,914) and the RUiE is 43% ( 1 / 2.31).

Still a usable metric, but one with where if I reached my optimal goal of 1.0 I’d still have some space left while I built (or moved to) the next Data Center.

Food for thought?  Comments welcome.

1 Comment »

Category: Data Centers Food for Thought     Tags: ,

1 response so far ↓

  • 1 shaloo shalini   November 19, 2012 at 1:48 am

    RUE concept is interesting and nicely tuned for say 90% of capacity (so you have more room if needed). Thansk for sharing. Here are some thoughts on expanding the concept even futher.

    What about floor space? You haven’t factored that in wrt capacity. Say filling up all floor space although only 40% of Us are actually in use? Or using only 10% of available floor space but paying on real estate costs for 100% space?

    Also, for HA you need to have some spares. Say if utilization is 40% on an average then how many spare Us should be there in RUE context?