Whatever their long-term view ought to be, many people see the
premium on green goods as discretionary, and among the first thing
to be cut when times get tough. To take just one example, UK
organic food sales are estimated to have slumped by a third in the
last six months. But for datacentres, green could be the colour of
survival.
Green storage is not just a tick in the box on the way to
meeting corporate social responsibility targets. The many different
technologies that fall under the
green storage heading can all be used to rein in the
datacentre's fastest-growing overheads, and can help postpone the
expansion or relocation of datacentre premises. It may even be the
only approach that will enable companies to keep up with the
ever-swelling flood of data, and thereby stay in business.
Last year Gartner warned that by the end of 2008, nearly 50% of
datacentres worldwide would be facing increasing difficulties in
finding the electricity they needed to power - and cool - their
equipment. Even if they can find the power, organisations which
currently spend 4-8% of their IT budgets on energy may find their
costs rising as much as fourfold within the next five years.
Gartner research vice-president Rakesh Kumar says that although
the threat still looms, people are finding ways of working around
it, such as
virtualisation, running datacentres at higher temperatures, and
using outside air for cooling. "Many of the datacentres in use
today were built 10 or 15 years ago and are not appropriate for
current requirements. People are building new sites, or going to
hosting players - demand for hosting is exceeding supply in major
cities in Europe and the US."
Gartner's figures are quoted in a 2007 report to Congress on
datacentre energy consumption by the US Environmental Protection
Agency (EPA), which adds that over the next five years, power
failures and limits on power availability will halt operations at
more than 90% of datacentres.
The EPA also reports that 40 out of 100 datacentre operators
said they had run out of space, power or cooling capacity without
sufficient notice. The likelihood is that it is data storage, as
much as processing, that has brought them to the brink of meltdown.
Consumers uploading videos, businesses storing multiple instances
of the same data for different applications and for different
levels of backup, and regulatory requirements such as Sarbanes
Oxley in the USA, are just a few of the sources of the surge. Some
commentators suggest our data storage requirements could soon be
doubling every year. Among soberer estimates, IDC puts the increase
at 60% a year and rising.
According to IDC's March 2008
white
paper "Driving Reduced Cost and Increased Return from the Data
Center" (prepared on behalf of Fujitsu-Siemens), electricity
currently accounts for 8.4% of datacentre running costs. IDC says
the same amount of energy is consumed to supply power and cooling
as to run IT equipment. "This means each watt saved in terms of IT
equipment counts double because one watt is also saved in terms of
the infrastructure." Other estimates say overall consumption is
around 2.5 times direct consumption by the equipment.
Most attention up to now has focused on servers, with
improvements to both energy efficiency and virtualisation. But
various industry estimates suggest we are coming to the limits of
the gains that server virtualisation can make.
According to the
Storage
Networking Industry Association's Green Storage tutorial, the
proportion of datacentre power and cooling costs associated with
storage can vary from less than 10% to more than than 40%. The
SNIA's rule of thumb is 60% for servers, and 20% each for
networking and storage.
But that 20% rule of thumb should be regarded as a starting
point only. Storage is not seeing the energy efficiencies which are
being achieved with other technologies. The EPA's report to
Congress says that while power consumption by networking equipment
and servers grew 14% and 13% respectively between 2000 and 2006,
storage was up by 20%.
However, the SNIA says there is "no consistent definition of
storage". If, as the old industry adage goes, you cannot manage
what you cannot measure, you are in an even tougher position if you
cannot even fully define what you are trying to bring under
control. The SNIA is working to establish metrics for storage
energy consumption and management. The
US Green Grid is working on a broader set of metrics for the whole
datacentre.
Once upon a time, when people were cheap and hardware was
expensive, managing disk to get highest rates of utilization and
performance was a prized art. But with storage getting cheaper by
half every year, according to Clod Barrera, chief technical
strategist at IBM Storage, people "stopped managing storage with
real science and real tools".
Instead, storage planning became a matter of deploying more than
you thought you would need, so you did not get caught out. Now,
Barrera says, even if you can afford the capital expense of more
disk, you cannot afford the power to run it, so "there is a
significant value-add for doing a better management job".
If we could parachute in teams of hardy veterans from mainframe
days, we could perhaps learn to do more with what we have already
got. But these are the people likeliest to have been squeezed out
in the 1990s because their skills were seen as obsolete and too
expensive, their frugality and obsession with efficiency and
"elegance" too narrow for an age of growth without limits.
Information Lifecycle Management is the modern equivalent of
those old skills. As IDC defines it, ILM "aligns requirements from
business processes (expressed in service level objectives) with the
possibilities offered by the available storage resources in an
optimized process. Policies control the dynamic processes if, for
example, the value of the information objects changes over
time".
Mission critical data can be stored on high performance
Serial Attached SCSI (Sas) disks, the 3Gb/sec replacement to
Ultra SCSI introduced in 2004. Recently introduced 2.5-inch Sas
disks claim 50% power and cooling efficiency over their 3.5-inch
predecessors, as well as taking up less space. Or you could simply
replace old hard drives new drives offering several times the
capacity will also be more energy efficient.
Less critical "nearline" data can be stored on "cheap and power
efficient" (IDC) Serial Advanced Technology Attachment (Sata)
disks, which are used in Massive Arrays of Idle Disks (Maid)
systems. Disks are only "spun-up" when the data on them needs to be
accessed. Maid is designed for maximum utilisation rates of 25% -
but this is at the top end of the range for all arrays, even those
in online use. Maid is designed for persistent or "write once, read
occasionally" data, as opposed to transactional data which is
updated and accessed frequently.
For archiving, tape consumes a fraction of the power of disk -
supplier and analyst claims range between 20 and 100 times more
energy efficient. Tape once played a major role in backup and
recovery perhaps it is time to rethink how much data needs to be on
"hot" standby on disk for the business to function, and how much
could be recovered with less urgency from tape. Users' and
customers' performance expectations may have to be managed
downwards if the recessionary climate deepens, these expectations
will probably fall anyway.
Storage (or Hierachical) Resource Management tools can help by
analysing usage patterns, and identifying under-used capacity and
data that could be moved to near or offline storage. SRM is also
used to predict future capacity needs.
An international
working group is defining standards for SRM web services.
Setting ILM policies is not as easy as it used to be. How do you
define mission-critical in these days of outward-facing,
entertainment-based services? Will you lose customers - or eyeballs
for the ads that provide your revenue - if people have to wait too
long for the skateboarding kitten video?
And how many copies of the skateboarding kitten video are you
storing? Data de-duplication can help. De-duplication specialist
Diligent Technology defines it as "a method for finding and
eliminating redundant data from the network and/or storage
infrastructure". A single instance of data appears to multiple
applications as multiple "nominal" instances.
In its white paper
Guidelines for the Evaluation of Enterprise De-Duplication
Solutions, Diligent explains, "the de-duplication ratio is the
ratio of nominal data to the physical storage used... A 10:1 ratio
means that 10 times more nominal data is managed by the system than
the physical space required to store it." Some suppliers use voodoo
calculations to claim ratios of up to 500:1. Diligent says 25:1 is
more realistic. Diligent was acquired by IBM earlier this year its
engine is used by HDS, Sun and Overland Data.
De-duplication can be oversold as a green technology. As IBM
Senior Storage consultant
Tony Pearson points out on his blog, a de-duplicated disk is
greener than a non-de-duplicated disk, "but not as green as
storing the data on non-de-duplicated tape".
Storage virtualisation is another approach. Gartner's Kumar says
there is a lot of misinformation about virtualisation the benefits
only come when you decommission the machines identified as
under-used. "It is good operational housekeeping, but it is not
going to make an enormous difference to power consumption." As for
using the redundant machines to cope with future demand, he points
out that you will lose out on the substantial energy efficiency
advances of newer technology.
"Thin provisioning" specialist 3Par claims virtualisation
requires "just one terabyte of capacity for every 2.5 terabytes
required with traditional storage arrays", and can cut storage
carbon footprints by 60%. Earlier this year 3Par announced that its
customers are responsible for "estimated worldwide combined
annualized energy savings of approximately £3.6m, equivalent to
eliminating approximately 48,000 metric tons of CO2 and 250,000
kilowatt-hours of electricity."
These are rare hard figures in the quest for green storage.
Without standardised metrics, datacentre managers who are
struggling to balance demands for growth and performance with
energy efficiency are at the mercy of supplier claims.
And unfortunately the benefits of green technology have been
over-hyped to the point where, according to a survey by Overland
Data, UK buyers are "incredibly sceptical" about these claims.
Overland also found only 22% of UK storage buyers put "green" in
their top three criteria, compared to three-quarters of buyers in
France and Germany. But a hard-headed business-based approach to
cost reduction through energy efficiency will reach the same goal
as an overtly green strategy it may even get there more
quickly.