The power of struggle
Rising energy prices and the threat of climate change are driving green issues to the top of the corporate agenda. Lynn Bairstow, Data Centre Products Director, Capgemini and John Mills, Data Centre Services Delivery Director, Capgemini review the potential for grid and utility computing to improve energy efficiency and sustainability in data centre management and procurement.
1 December 2006
Publication
This finding was further underlined by additional research the firm conducted with CIOs at six of the UK’s largest financial services companies – once again power was considered to be the number one problem by each of these organisations. The research found that while only the largest data centres experienced significant problems in 2005 it is anticipated that more data centres are going to come under increasing pressure in the future as administrators continue to replenish older equipment with newer units that have higher power densities.
The fact that power should come out on top is unsurprising given the current environment – external factors such as rising energy prices, increased competition for fossil fuels and the threat of global climate change have all driven sustainability and green issues to the top of organisations’ agendas. As a result, companies are now responding with policies intended to curb their use of energy and review the sustainability of their IT infrastructures. The issue is particularly critical within the data centre. Increasing power densities of networked storage, communications equipment and servers all contribute to a spiralling demand for energy to power and cool the data centre. Servers present corporations with one of the biggest headaches because server farms have grown so large.
In fact an average data centre will typically use 55 times more energy than an
average office block. To put this into context, an average office block operating
for 10 hours a day might consume 100 watts per square metre in those 10 hours.
A data centre uses approximately 1,000 watts per square metre and operates 24
hours a day, seven days a week. In addition to the energy used to power the servers,
there is also the energy required to cool the data centres. For every 1 kilowatt
per square metre of energy required to power a data centre, an additional 0.7
kilowatt per square metre of energy is required for cooling. This equates to a
70% surcharge to deliver the required extra 1000 watts square metre of energy
in the data centre. It is vital therefore that data centre managers focus on
efficiencies, a 1600 square metre data centre can face power bills in the region
of £700,000 per year so even a 1% saving is significant.
The challenges
Increasing circuit density and processor power has led to greater energy consumption and associated heat output. High density technology solutions, such as blades, provide significantly improved CPU power per square metre ratio, however, they do still have the downside of heat dissipation issues and usually require specially designed data centre facilities to cater for the concentration of heat, power and floor loading.
Manufacturers have recognised the cooling challenge that greater circuit density presents; hot and cold aisles, adjusting floor tile ventilation and other conventional strategies are being overtaken. Innovations are therefore being developed such as intelligent use of component circuitry; turning off unused elements within a processor; rows of rack-based water cooling solutions and direct expansion refrigeration units mounted within racks. Other actions such as greater use of fresh air circulation overnight and ensuring in the winter all elements are kept filtered and cleaned, helps to reduce the reliance on power input and keeps UPS and other equipment cool. Potentially an ice storage system could contribute to the savings by being charged up at night when the efficiencies are there and discharged over peak periods during the day.
Better use of technologies that have improved carbon footprint for energy generation such as on-site heat exchangers distributing waste heat for third-party use, and the use of solar and turbine alternative power generation, can also play a key role in helping to tackle the power problem in data centres.
However, it is apparent that though some efficiency will be produced by more efficient servers, energy-efficient chips and cooling systems, such measures will only go some way to compensate for the increasing data processing requirements of companies. They are not enough in themselves to alleviate the ongoing issue that organisations face in their overall data centre energy consumption.
Is grid computing the answer to the data centre energy challenge?
Moving to a utility or grid computing model provides significant opportunities
for increased IT availability, pay-as-you-use billing and a greater return on
IT investment. The infrastructure is shared across the enterprise to smooth the
peaks and troughs in IT demand, allowing further external resources to be temporarily
tapped as required. Shared servers mean less physical servers and hence less energy
consumption. Data centre management still has a major responsibility to encourage
clients to utilise the highest efficiency boxes, and ensure that supporting plant
is operating at its highest efficiencies.
Sharing the infrastructure is achieved by virtualising the infrastructure (viewing,
managing and scaling the infrastructure as pools of computing resource) and dynamically
provisioning it (using significant degrees of policy-based automation). Under
the utility model servers are directly provisioned to an application, whereas
the grid model uses applications that ‘find’ spare server cycles across one or
more resource pools. Dynamic billing can then be applied to these models on a
pay-per-use basis, but is not a requirement of the shared infrastructure.
However, it is a complex process and needs to be managed carefully to get the best results. The grid model requires implementation of several enabling toolsets and a management approach to provide pooled capacity across natural boundaries (internal, geographic and organisational). Given the scale of change involved (new management approach, technology implementation and skills searching), an industry-wide move to a utility or grid environment is not likely to be the quick answer. In the short term, many organisations will likely take some first steps towards one or more of the virtualisation, dynamic provisioning or pay-per-use billing components of a shared environment; leaning on the collaborative efforts of internal and external organisations to ease the process.
While fully shared infrastructures are not realistic now (and may never be due
to security and other constraints) there is value in collaborating with strategic
partners to make steps in this direction and with whom companies can share risks.
In practice moving to a utility or grid computing model requires significant transformation
to overcome technical and organisational barriers. Collaborative partners and
outsourcers for example are well-placed to manage such transformations, aggregating
complex services, providing access to technical expertise and state-of- the-art
infrastructures, and provision cost savings through economies of scale.
Conclusion
Ultimately, the return on investment for utility or grid computing will be on
a sliding scale; a higher priority for businesses with high degrees of volatility
and frequent IT infrastructure change, but not for others with flat IT usage and
infrastructure change profiles. It is clear then that a fundamental shift needs
to take place around how organisations procure and manage their data centres to
increase energy efficiency and sustainability. Grid computing, virtualisation
and provisioning on-demand can aid energy efficiency as well as reduce physical
infrastructure requirements. However, when entering into collaborations with suppliers
and managed data centre service providers, companies must also ensure that one
of the key selection criteria they use is sustainability.
To read the article, click here.
