Calculating data center energy costs and boosting energy efficiency
By Carol Ko 07-Aug-2012
Assuming you are a data center operator in Hong Kong running 2,000 racks of servers round the clock, your energy bill is likely to charge you some US$12 million each year.
The estimation comes from Gabriel Leung, Hong Kong Computer Society's director of community services and EMC Hong Kong & Macau's general manger. He worked out the power costs based on the actual figures of Asia Tone, a Hong Kong-based data center operator that was acquired by global data center services provider Equinix last month.
"[Y]ou are likely to pay about HK$100 million (US$12.89 million) to power up a 2,000-rack data center in Hong Kong for one year."
-- Gabriel Leung, director of community services, Hong Kong Computer Society
According to Leung, each server rack, when taking into account of cooling and redundant power usage, consumes about 3kVA (aka three units of power) per hour. "Given each rack consumes 3 power units per hour, the power consumption for a year will amount to 26,280 units of power (3 power units x 24 hours x 365 days). Assuming each data center houses 2,000 racks, the entire data center will consume up to 52 million units of power in a year (2,000 racks x 26,280 power units = 52,560,000 power units/year)," Leung said.
"And assuming your power company CLP charges HK$1.5 per power unit, your energy bill for running a data center in Hong Kong for one whole year will be HK$78,840,000 (HK$1.5 x 52,560,000 power units). If you take into account of other utility expenses such as lighting, you are likely to pay about HK$100 million (US$12.89 million) to power up a data center for one year," he added.
Energy efficiency matters
Considering an annual energy bill of US$12 million, it is important for a cloud data center to be energy efficient.
"We are increasingly being told by cloud computing customers that energy efficiency is being incorporated into cloud computing contracts, for example in terms of PUE [Power usage effectiveness, a metric used to determine the energy efficiency of a data center]," said Peter Sharples (pictured above), director of channel strategy, cloud and IT management at CA Technologies.
According to Sharples, energy-related costs and characteristics are increasingly being used as a differentiator between cloud providers and service providers. In particular, this will help increase the efficiency of the infrastructure, resulting in lower costs being passed on, as well as helping with greater reliability through increased insight and more effective management.
"Energy efficiency is one goal, but it exists alongside other goals such as reliability, agility, cost-effectiveness, effective use of capacity, security, and so on. The relative importance of energy efficiency will differ depending on a number of factors," he said.
In an interview with Asia Cloud Forum's Carol Ko, CA Technologies' Peter Sharples explains the difference in energy efficiency between a cloud data center and a traditional one, comments on the possibility of powering cloud data centers with renewable energy sources, and tells us if CA powers its data centers using renewable energy sources. Below are the excerpts of the interview:
Asia Cloud Forum: How does a cloud data center differ from a traditional data center in terms of energy efficiency?
Peter Sharples: Consolidation, standardization, virtualization and centralization have the potential to be more efficient, but not necessarily so. We recommend DCIM [short for Data center infrastructure management, which refers to the use of IT and facility management disciplines to centralize monitoring, management and intelligent capacity planning of a data center's critical systems.] to enable cloud providers to have insight into efficiency and to achieve more professional monitoring and management of their critical infrastructure.
What are the metrics of measuring the energy efficiency level of a cloud data center?
Sharples: PUE is the most frequent metric we encounter. This metric has its proponents and detractors, but we do see it very commonly used.
We are also increasingly seeing water-related and greenhouse gas emissions-related efficiency metrics, such as CUE (Carbon Usage Effectiveness).
We see chargeback metrics commonly used by service providers. This includes energy consumption, energy chargeback cost and also, in some cases, peak power demand.
Other subsidiary operational metrics relate to energy intensity, percentage virtualized and so on. What we have not seen much of yet, but what we expect to expand considerably are a series of metrics that relate to power per virtual machine or power associated with a specific business service.
We are seeing metrics evolve among some of our customers concerning energy per customer (or energy normalized to high-level business metrics).