About
Subscribe
  • Home
  • /
  • TechForum
  • /
  • Fuel costs and high density computing loads - an analogy

Fuel costs and high density computing loads - an analogy

Johannesburg, 31 Mar 2006

Analogies are often the best way to learn about complex issues. This can be especially relevant if you are responsible for specifying the power and cooling capabilities of a data centre to support high density IT equipment. Here the analogy relates the specification of a car`s fuel efficiency to that of a data centre`s power density.

Assume that you are in the market for a new car and need to specify its fuel efficiency given the exorbitant cost of petrol. Would you specify kilometres per tank? Describing fuel efficiency using kilometres per tank is not sufficient to determine the capability of that car because the tank capacity varies from model to model. Instead, you are likely to specify kilometres per litre of petrol because value is comparable across all car makes and models.

Similarly, why would you specify a data centre`s power density using Watts per square meter (W / m2) as is most often the case? What is included in area and what is included in power? The problem with this description is less obvious than the "kilometres / tank" example, but is nonetheless ambiguous and misleading and is not sufficient to determine power or cooling compatibility with high density computing loads like blade servers. The sidebar "Density Specification Methods" clearly shows that density specifications for the same facility can vary by almost a factor of eight depending on the density definition used. Yet, this is exactly how the power density for most, if not all, data centres is specified today. This has led to confusion in the industry and miscommunication between IT personnel and facilities designers and planners. The historical method of specifying data centre density in Watts / m2 provides little guidance for answering critical questions that are faced by data centre operators today. In particular, the historical power density specification does not answer the key question: What happens when a rack is deployed that exceeds the density specification? This is a very practical question because the typical data centre has a density rating of 1.5 kW per rack while modern IT equipment has a greater power density of 3-20 kW per rack.

Real data centres do not exhibit a uniform power density. Some racks draw more electrical power and consequently generate more heat than others. Patch panel racks may draw zero power. Blade server racks may draw 20 kW or more power. Compounding this problem is the fact that IT equipment is constantly being refreshed, which means the power consumption of particular racks is subject to change over time. Conventional density specifications don`t fully take in to account these power variations and as time goes on they become less effective. American Power Conversion (APC) proposes that Watts / rack be used as the basis for an improved data centre density specification method. This specification method assures compatibility with anticipated high density loads, provides unambiguous instruction for design and installation of power and cooling equipment, prevents oversizing and maximises electrical efficiency. Specifically, this method meets the following requirements:

Predictability: Permits the ability to determine the power and cooling capacity at any rack location for any proposed or actual installation of IT equipment. Accept partially specified future requirements: Requires that the exact power be known in advance for each rack location. In fact, IT equipment only lasts for a fraction of the life of a data centre and is routinely changed out for new and different equipment.

Support power and cooling borrowing: Available power and cooling that is not used at a specific rack should be available for use by other racks.

Minimise waste: Electrical inefficiency should be minimised. Available power, cooling, and space should be utilised. Capital and operating costs should be minimised.

Support staged deployment: Supports a staged deployment, including the case where different stages may be at different densities, and where the data for future deployment stages is not known at the time of earlier deployments. With these requirements in mind, a model can be created for density specification, which provides many key benefits including:

* It provides a more complete and accurate description of data centre density than other commonly used specification methods;
* Data centres built to the specification will have more predictable performance;
* The model is specific enough that costs, including capital and operating costs, can be rapidly estimated, speeding the design cycle and permitting alternate scenario analysis; and
* It supports a system of modular scalable data centre deployment, which can dramatically reduce TCO and improve electrical efficiency.

Practical applications of the described density specification method include:

* Comparing TCO associated with alternate data centre sites or room locations;
* Estimating costs associated with increasing density in a planned or existing data centre; and
* Providing a specification that clearly establishes density expectations in a form comprehensible to IT users, so that IT users, data centre operators and data centre systems suppliers establish the same expectations.

Sidebar - Density Specification Methods

The definition of power density is inconsistent in the literature, which has resulted in significant confusion in the user communities. In order to better understand these definitions, consider the following hypothetical 500 kW data centre:

The table, click here, shows five different commonly used definitions for power density and the values that result when applied to the data centre described above.

The clearest statement of density is the "per rack" power consumption. This provides unambiguous guidance regarding the power and cooling requirements of a rack (for IT equipment, the rack`s electrical power consumption in Watts equals the cooling requirement in Watts). Per-rack power consumption has another major advantage in the specification of data centre density - namely, that it is the most effective way to specify variations of density within a data centre.

Share

Editorial contacts

Jacqui Gradwell
APC by Schneider Electric
(011) 465 5414
jacqui.gradwell@apc.com