IT directors are being forced to focus on the
management and structure of their datacentres, as it becomes
clear that there is not a limitless supply of energy. Plugging yet
another server into the datacentre is no longer an option for
companies operating in Canary Wharf, where companies are facing
major difficulties securing extra power.
Increasing the efficiency of a datacentre has become a
necessity for many CIOs, and the most effective measures often
involve housekeeping and optimisation to see change in the short
term.
Although some companies are currently unaffected by power
shortages, the cost of powering and cooling a datacentre is
becoming a pressing concern. A
study published
last year by consultancy BroadGroup found that the average
energy bill to run a corporate datacentre in the UK is about £5.3m
per year. The report predicted that this would double to £11m over
five years and that the UK would become the most expensive place in
Europe to run a datacentre.
For the datacentre "super-user", unlike smaller businesses,
access to energy is an issue.
Google and Microsoft
are building datacentres the size of double football pitches in
the southwest of America, where hydro power is plentiful. But
Tikiri Wanduragala, senior server consultant at IBM, says, "Every
large UK account is also talking about power consumption of
datacentres. It is their top priority."
Mike West, managing director of datacentre specialist Keysource,
says, "Medium-sized businesses are also having problems removing
heat with their legacy cooling systems."
Organisations have arrived at this juncture partly because
manufacturers and end-users have been working at cross-purposes.
"For a long time manufacturers have been obsessed with occupying
less physical space and producing
ever-smaller chips and boxes," says David Elwen, a director at
IT consultancy DMW Group.
One consequence of this is that organisations are faced with
having to find ways to
power and cool denser and hotter components crammed into the
same amount of space.
Datacentre designers have also been guilty of "comfort cooling"
for people rather than computers.
Google published
research in February 2007 that found disc drives can tolerate
temperatures of 38 degrees Celsius without suffering a higher
failure rate. The findings, based on five years' study of Google's
own datacentres, contradict established thinking that
equipment must be kept cool to function.
Compounding these design follies is a lack of communication
between facilities managers and IT managers. "The facilities
manager is a very different animal to the IT manager. He knows
about power loads and distribution of airflows, but nothing about
the IT equipment that sits in the cabinets he is looking after,"
says Elwen.
Hassan Moezzi, director of business development at datacentre
design and operations specialist Future Facilities, says, "IT staff
are ignorant of the physics of cooling, and IT departments are
shifting kit in and out of datacentres like a yo-yo".
The problem is exacerbated by the fact that the key aspects of
the datacentre - management and usage - are often outsourced to
different parties, and so the two parts remain distinct.
The separation between energy consumption on one hand, and IT
procurement and usage on the other, was further confirmed last
month by a Green Technology Initiative survey. It found that only
20% of user organisations surveyed considered both IT requirements
and energy consumption when purchasing equipment.
Dan Sutherland, founder of the Green Technology Initiative,
believes that making IT directors accountable for the
infrastructure they deploy and manage would make an immediate
difference to carbon emissions.
"If IT directors were targeted and given budgets and bonuses
according to how much they spent on power, they would look a lot
more carefully at the kit they use in the datacentre. Utility bills
have for so long been merely treated as an accepted annoyance,"
says Sutherland.
Getting people from all corners of the IT operation around one
table is the new challenge, says Wanduragala. "It is a big
challenge and an even bigger struggle. It requires the
communications manager, the storage and the server guys and the
procurement people to all sit down together."
Wanduragala says he is witnessing a shift in datacentre
management away from software and server teams towards the
datacentre manager. "Now any procurement decision must be based on
all parameters, and the biggest of those is raw efficiency," he
says.
The smart action for the IT director is to stop the problem
getting worse, says Wanduragala. This could entail a veto of new
hardware purchases until existing capacity has been exploited. "You
may have to make people jump through hoops before they get to buy a
new server," he says.
Such a course of action would probably mean virtualising every
server, which, in turn, means persuading business units to share
servers. Wanduragala says he is observing an alternative trend of
organisations replacing multiple servers with one
higher-performance - and more efficient - server.
Virtualisation is one formula for squeezing every ounce of
performance out of existing infrastructure. Another option that
would work for power-intensive applications is
grid computing. Ian Osborne, project manager at Grid Computing
Now, says, "The datacentre of the future will look like a much
more flexible infrastructure."
Osborne says that grid middleware could allow the IT manager to
distribute a processing task across available resources. With idle
resources continuing to consume 85% of power, grid computing is an
attractive option, although it is only really an option to CIOs
managing huge scientific and financial tasks.
Turning applications off when they are no longer needed
constitutes a big, and more immediately accessible, saving. An
audit of the applications of a blue chip company by DMW
Group revealed that 20% were never used, yet remained switched
on.
"The problem is that when replacement systems are introduced no
one goes around switching legacy applications off," says Elwen. He
recommends putting redundant applications into a "graveyard" area,
where they are eventually archived and switched off.
Virtualisation, grid computing and better house-keeping are all
options within the IT manager's domain, but just knowing what you
consume is a good start.
West says, "IT managers are surprised when we do an audit
comparing the consumption of the computer room with the rest of the
building. It is often three or four times greater than expected and
can account for 25% of the output."
Powering and cooling your datacentre
How significant is the cost of powering and cooling a
datacentre?
Ten years ago, 70% of the total cost of acquisition and
management of servers was in the cost of the hardware. Today, about
70% of the cost of managing servers revolves around power and
cooling, according to
research from analyst firm IDC.
Are there alternative sources of power?
Unlike the US, where super-consumers are attracted to states
with the most attractive power supplies and local laws, UK users
are more or less constrained to the National Grid.
Purchasing alternative sources of energy has debateable value,
says David Elwen, director of DMW. "Frankly, you can forget about
renewable energy in the UK. It is tokenism."
However, appealing the wind or alternative energy source is,
simply poured into the grid, and could end up at any destination,
Elwen says.
"The only way to use truly renewable energy is to produce it
yourself and then sell any surplus back to the grid. That is the
reverse of the usual datacentre that runs on mains and generates
its own back-up," says Elwen.
Some hospitals produce their own power, but the exact science of
generating at a constant load and strict planning permission
requirements make it an unappealing option for datacentres.
What are computer manufacturers doing to lessen power
consumption?
Choice of power supply may be largely out of the hands of UK IT
directors. Likewise, the physical architecture of chip components
and servers, and its consequent draw on power, is a given, although
efforts are being made by manufacturers.
Earlier this year,
IBM announced a breakthrough in the design of chips from the
traditional 2D design to a 3D stack. IBM's Semiconductor Research
and Development Center has found that stacking multiple chips
together uses less power and generates less heat.
Is cooling datacentres a big issue?
"Every megawatt required to power hardware takes another
1.5Mwatts to cool it," says Hassan Moezzi, director of business
development at Future Facilities. Choosing better cooling methods
is very much within the scope of the IT manager, and it makes a big
difference to carbon emissions.
Also, laying out racks and cabinets randomly can interrupt
complex and cooling airflows and create hotspots.
Future Facilities advocates that, instead of aiming for a
blanket temperature across a room, IT directors consider the
datacentre as a giant waterbed. "If you press one end, the other
will be affected," he says.
However, software that simulates the airflows around a
datacentre to help the IT better plan the layout of kit is
available.
Is there alternative cooling?
Some firms are selecting under-floor refrigeration methods
instead of air conditioning. This allows for more localised
cooling.
Other companies are seeking alternatives to refrigeration,
exploring options such as ground water and fresh air to cool racks
and cabinets of computing gear.
Rackspace to open green datacentre >>
Metrics needed for green datacentres, says expert
The Google
disc drive tolerance research >>
Aperture research paper: datacentres turn to high-density computing
>>
University of California: refrigeration assisted spot cooling of a
datacentre >>
Have your say
What is your take on Helen Beckett's opinion? E-mail
computer.weekly@rbi.co.uk