Unit 3: Business/IS Strategy and Planning
Seminar: Improving Energy Efficiency
We continue our study of business/IS strategy and planning, using the European Union data centre operations as an example. These principles and practices can be applied to companies as well as government agencies.
EU Code of Conduct for Data Centres
The European Commission, the executive arm of the European Union, issued Version 1 of a European Code of Conduct for Data Centres Energy Efficiency and Best Practice Guidelines on October 30, 2008. (See http://iet.jrc.ec.europa.eu/energyefficiency/ict-codes-conduct/data-centres-energy-efficiency for updates.) These are intended to reduce and assess the reduction of energy consumption in data centres. The Code of Conduct is voluntary.
The Code of Conduct argues that data centres have typically been designed with large tolerances for operational and capacity changes, and this results in power consumption inefficiencies. Only a small fraction of the power consumed by the data centre is actually used by the IT systems; most energy is used up by redundant power and cooling systems.
Focus of the Code of Conduct
- IT load: consumption of energy of the IT equipment in the data centre
- Facilities load: air conditioning/cooling units for mechanical and electrical systems (chiller plant, fans, pumps)
The Code of Conduct considers the data centre as a complete system. The goal, then, is to optimize the IT system and the infrastructure together to achieve facility efficiency.
Data Centre Infrastructure Efficiency Measure
The code initially defined one metric: Data centre infrastructure efficiency (DCiE). DCiE is equal to the main IT equipment energy consumption divided by the total facility energy consumption (the range is between 0 and 1). A higher figure indicates better energy efficiency. A measure of 0.5 would indicate that half the energy consumed by the data centre was used by the IT systems (a typical figure for existing centres). A theoretical figure of 1 would indicate that all the energy consumed by the data centre was used for IT (unlikely).
Further metrics developed include an IT productivity metric and a total energy productivity metric. The IT productivity metric indicates how efficiently the IT equipment provides useful IT services. This requires some measure of output of the centre, such as the number of standard tractions performed per unit of energy. The total energy productivity metric relates the useful IT services to the total energy consumption of facility.
Best Practices for Data Centres
The updated (2014) Best Practices are a companion to the EU Code of Conduct for data centres energy efficiency to assist with measures to improve energy efficiency. All expected practices should be applied to any data centre built from 2011 onwards.
Expected Minimum Practices
The majority of the best practices are now defined as the expected minimum level of energy-saving activity. Numeric values of 1 to 5 indicating the effect of the measure are included. The general categories are as follows:
Data Centre Utilization, Management and Planning
Involvement of organizational groups: Ineffective communication between the disciplines working in the data centre is a major driver of inefficiency as well as capacity and reliability issues.
General policies: These policies apply to all aspects of the data centre and its operation.
Resilience level and provisioning: One of the most significant sources of inefficiency in data centres is the over-provisioning of space, power, or cooling and the facilities being run at part capacity. Monolithic, as opposed to modular design of facilities also represents a significant and frequently unnecessary capital expenditure. Further, as the level of resilience of the data centre increases, the inefficiencies due to fixed overheads increase, and this is compounded by poor utilization.
IT Equipment and Services
Selection and deployment of new IT equipment: Once IT equipment is purchased and installed in the data centre it typically spends several years in the
data centre consuming power and creating heat. The appropriate selection of hardware and deployment methods can provide significant long-term savings.
Deployment of new IT services: The service architecture, software, and deployment of IT services have an impact at least as great as that of the IT hardware.
Management of existing IT equipment and services: It is common to focus on new services and equipment being installed into the data centre but there are also substantial opportunities to achieve energy and cost reductions from within the existing service and physical estate.
Data management: Storage is a major growth area in both cost and energy consumption within the data centre. It is generally recognized that a significant proportion of the data stored is either unnecessary or duplicated nor requires high performance access and that this represents an organizational
challenge. Some sectors have a particular issue due to very broad and non specific data retention directions from governments or regulating bodies. Where there is little structure to the data storage, implementation of these regulations can cause large volumes of data not required by the regulations to be unnecessarily heavily protected and archived.
Air flow management and design: The objective of air flow management is to minimize bypass air, which returns to the cooling (CRAC/CRAH) units without performing cooling, and the resultant recirculation and mixing of cool and hot air, increasing equipment intake temperatures. To compensate, cooling unit air supply temperatures are frequently reduced or air flow volumes increased, which has an energy penalty. Addressing these issues will deliver more uniform equipment inlet temperatures and allow set points to be increased (with the associated energy savings) without the risk of equipment overheating. Implementation of air management actions alone does not result in an energy saving – they are enablers which need to be tackled before set points can be raised.
Cooling management: The data centre is not a static system and the cooling systems should be tuned in response to changes in the facility thermal load.
Temperature and humidity settings: Facilities are often overcooled with air temperatures (and hence chilled water temperatures, where used) colder than necessary resulting in an energy penalty. Widening the set range for humidity can substantially reduce humidifier loads. Reviewing and addressing air management issues is required before set points can be changed. In order to avoid risk to operational continuity, expert advice should be sought before changing the environmental range for the facility. An increase in chilled water temperature set points provides enhanced efficiency for free cooling economizers and a reduction in compressor energy consumption. Unnecessary humidifier loads generated by chilled water or evaporator temperatures below the data hall air dew point causing dehumidification should be eliminated through adjustment of the lower humidity set point.
The cooling plant typically represents the major part of the energy used in the cooling system. This is also the area with the greatest variation in technologies.
High-efficiency cooling plant: When refrigeration is used as part of the cooling system design high-efficiency cooling plant should be selected. Designs should operate efficiently at system level and employ efficient components. This demands an effective control strategy which optimizes efficient operation, without compromising reliability.
Even in designs where the refrigeration is expected to run for very few hours per year the cost savings in infrastructure electrical capacity and utility power availability or peak demand fees justify the selection of high-efficiency plant.
Computer room air conditioners: The second major component of most cooling systems is the air conditioner units within the computer room. The computer room side of the cooling system is frequently poorly designed and poorly optimized in older facilities.
Data Centre Power Equipment
Selection and deployment of new power equipment: Power delivery equipment has a substantial impact upon the efficiency of the data centre and tends to stay in operation for many years once installed. Careful selection of the power equipment at design time can deliver substantial savings through the lifetime of the facility.
Other Data Centre Equipment
General practices: These general practices apply to the data floor and may be extended to the remainder of the building if no sustainable building standard is in use.
Energy use and environmental measurement: Most data centres currently have little or no energy use or environmental measurement capability; many do not even have a separate utility meter or bill. The ability to measure energy use and factors impacting energy use is a prerequisite to identifying and justifying improvements. It should also be noted that measurement and reporting of a parameter may also include alarms and exceptions if that parameter passes outside of the acceptable or expected operating range.
Energy use and environmental collection and logging: Once data on energy use and environmental (temperature and humidity) conditions is available through the installation of measurement devices it needs to be collected and logged.
Energy use and environmental reporting: Energy use and environmental (temperature and humidity) data needs to be reported to be of use in managing the energy efficiency of the facility.
McKinsey & Company on Data Centre Efficiency
In Revolutionizing Data Centre Energy Efficiency (by J.M. Kaplan, W. Forrest, and N. Kindler, July 2008), McKinsey & Company recommend the appointment of an energy czar and the transfer of financial accountability of data centre assets from corporate real estate to the CIO (chief information officer). These recommendations are driven by an economic analysis that concludes that ongoing server proliferation is increasing the expenses related to power and cooling, which crowds out other IT initiatives unless IT gets a bigger slice of corporate revenue.
Some of the report’s findings:
- Data centres (DC) make up one quarter of total ICT costs, and rapid growth in the number and size of data centres creates two problems:
- Large cost: both capital expenditure (CapEx) and operational expenditure (OpEx) for data centres are large, quickly growing portions of total IT budget.
- Large CO2e: For many organizations, data centres are the largest source of greenhouse gas emissions.
- Causes for inefficiencies:
- Poor planning for demand and capacity,
- Asset management failings, and
- Energy disconnect: boards, CEOs, and CFOs are not holding the CIO accountable for energy use or greenhouse gas emissions.
- Solutions for inefficiencies:
- Integrate asset management (as is done for corporate security),
- Mandate true total cost of ownership (TCO), and
- Appoint CIOs and internal “energy czars” (technology and operations mandate) to double energy efficiency by 2012.
- Metrics to measure efficiencies:
- The study proposes the Corporate Average Data Efficiency (CADE) metric for data centres across the corporate footprint.
- The study proposes that industry groups dialog with regulators to establish a benchmark or metrics to measure the individual and combined energy efficiency of corporate, public sector, and third-party-hosted data centres (for example, a similar benchmark is already implemented in the US automotive industry called CAFE: Corporate Average Fuel Economy).
Green Grid on the Use of Proxies
Using common indices such as the DCE (data centre efficiency) and the PUE (power usage effectiveness) can be more difficult than it first appears due to the problem of obtaining operational data from data centres constructed to provide continuous computing services rather than accurate energy usage reports. To overcome some of these complexities and to provide a first level of information on initial energy-reducing strategies, the Green Grid has proposed the use of a simple indicator or proxy.
Use of a proxy is understood to be less accurate than direct measurements such as DCeP (data centre energy productivity); nonetheless, proxies are useful because they are easier to implement than direct measurements, and they often provide the only information achievable within data centre structures.
The key objective is to find a proxy that can both substitute for the direct metric and provide a useful measure of the activity under observation. An example may be the use of Canadian Office of Energy Efficiency (OEE) metrics to provide passenger vehicle fuel consumption ratings, understanding that these ratings are produced under a specific set of circumstances; actual fuel efficiency will vary when vehicles are driven in different conditions.
Green Grid is engaging industry stakeholders to test a range of proxies to determine which features and techniques prove effective in applied use. Effective proxies can then be used to augment existing metrics such as the PUE. These proxies will provide data centre operators greater flexibility and easier-to-use tools to measure and monitor their activities. Such steps will facilitate the ability to take action in carbon reduction and the ability to meaningfully record the evidence of such action.
The Green Grid uses the following criteria to evaluate a useful proxy:
- ease of use
- time to implement
- operational ability
The Green Grid publication Proxy Proposals for Measuring Data Center Productivity lists eight proxies, describing the pros and cons of each and a possible use case for each.
To improve performance, it is essential to measure it. While the work is not complete, the use of proxies is both necessary and useful, as they provide data centre operators a range of usable tools to evaluate their centre’s performance. Where best practice is not achievable, an alternative needs to be found, particularly where urgent action is required and where time will not allow the implementation of best practice measurements.
Proxies can therefore play a key role in improving data centre performance by providing economical, efficient, effective measuring tools that do not interfere with the primary computing function of a data centre, but that provide usable data for both cross-sectional and longitudinal comparative analysis. Also, use of proxies provides a basis for other potential proxies that data centre owners may wish to use in their particular configurations.