ASHRAE Journal - June 2014 - 83

30
25
20
15

Servers & Disk Storage Systems

10
5
0

1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014

Figure 1: Thermal Management Consortium 1995 trend chart (filtered and
reformatted).

The evaluation of any manufactured product is a
multi-dimensional analysis that can include variables such as:
* Maximum IT and power demand;
* IT and power demand at no load;
* Load performance curve;
* Compute performance per watt;
* Form factor (1U/2U servers, switches, routers,
etc.); and
* Power, cooling, and network interfaces.
In evaluating the accuracy of the previously published trends, one method is to consider whether
at least one product per year for each product type
curve is close to what the trend curve forecasts. The
other critical factor is the determination that there
not be a single product that greatly exceeds the
trend curve in a given year.
A filtered and reformatted version of the 1995 version of the Power Trends Chart is shown in Figure
1. There is a single curve for Servers & Disk Storage
Systems. Having a single product category of Servers
& Disk Storage Systems made the values very generalized and all-encompassing.
This meant many products within that category
would fall way below the curve (e.g., storage servers), but that did not mean that the chart was inaccurate. Again, the curves represented the maximum
performance or worst-case scenarios.
For example, considering the value of a 6 kW/rack
in 2002 and using 42 rack units or "U" indicates that

Compute Servers - 1U, Blade & Custom

35
30
25

Compute Servers - 2U & Greater

20
15

Storage Servers

10
5
0

1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014

Figure 2: ASHRAe TC 9.9 2005 trend chart (filtered and reformatted).

the maximum 1U server load would be 140 W (i.e., 6
kW/42 servers). This would not be a value that would
be unreasonable for the best performing 1U compute
servers at the time, but certainly not a realistic value
for storage servers. The net result was intuitively
the curve represented a kW/rack that has never
been seen and that data center professionals cannot
imagine ever happening.
The 2005 version of the Power Trends Chart is
shown in Figure 2 with the same filtering and reformatting applied for comparative purposes. The
single product category of Servers & Disk Storage
Systems became three product categories with three
separate curves. This reduced the generalization by
having three different maximum kW/rack values
(one per category). The three product categories
are:
* Storage servers;
* Compute servers: 2U and greater; and
* Compute servers: 1U, blade servers, and custom.
Intuitively, it makes sense that the maximum kW/
rack would be different for these three product
categories. Anything that can be extracted from the
curves that most data center professionals would
agree upon without hesitation helps build confidence about the general accuracy of the curves. The
following are statements about the curve patterns
that should make intuitive sense to data center
professionals:
J u n e 2014

ashrae.org

ASHRAE JouRnAl

SouRCe: ASHRAe. ModiFied By dLB

Heat Load in kW/Rack (Based on Rack Footprint=7 ft2)

SouRCe: upTiMe inSTiTuTe. ModiFied By dLB

Heat Load in kW/Rack (Based on Rack Footprint=7 ft2)

Column Data centers

ASHRAE Journal - June 2014

Table of Contents for the Digital Edition of ASHRAE Journal - June 2014