Green
Moores Law states
that the number of transistors on a microprocessor will double about every two
years. The law has held up pretty much since Gordon Moore, one of the founders
of Intel, first published his paper on the subject in 1965. With the increase
in chip capacity came a rise in speed and therefore processing power. We have
powerful computers today because of the technology and innovation that drives
chip design and production. When the first men landed on the moon it sometimes
said that there was less computing power on the spacecraft than there is on a
modern mobile phone. So we’ve come a long way in 40 years or so.
For the
stand-alone home or office computer or laptop, the amount of energy consumed is
large in itself. Now consider taking an individual computer with all it’s
associated processor technology and less the peripherals, such as the screen
and mouse, and multiplying that by hundreds, maybe thousands of similar devices
in the same room. These rooms are euphemistically termed an ’server farm’,
 where lines upon lines of individual servers get stacked together to process
information. This scenario presents processor designers at the front end and
building services engineers at the back end, with the same problem, how to
dissipate heat and minimize power requirements. The reason server farms exist
in the first place is because our world is becoming more data-driven. And in
the world of 24/7 data requirements then the server farm is indeed a practical
solution.
Grouped servers
or server farms generate huge amounts of heat and because the servers must be
kept cool relative to their operating limits then a huge amount of energy must
be expended on ventilation and air conditioning. As the energy demand goes up,
so too does the cost. And because more and more companies are using these
server farms to process and warehouse data, then the demand for both the faster
technology and energy is rising in parallel. As the world is increasingly
becoming more speed and data-driven, increasing data requirements are driving
demand for more server capacity and therefore larger and more complex storage
locations.
An interesting
read, but a few things come to mind around the green IT space, firstly we need
to move the applications to what I call the BIG THREE, Web, Citrix/application
streaming and Grid/DataSynapse/Platform, etc. If they aren’t one of these three
media, then the possibilities around them are going to limit what we can
achieve (excluding the database of course). By that I mean, if I have a
proprietary application, which cannot for whatever reason be upgraded to a web
platform, be streamed in a Citrix or online java type application, or have the
workload converted into a grid type application, we will need to maintain the
server, the switch, the storage for the individual application server nodes.
What we need to do is:
Tier the
application – how available does the application need to be and to what
performance levels – if it’s a train service status in a developing economy
that runs three times a day and is accessed by mobile phones, does that need
the same level of service as the same application in New York accessed by
thousands of iPhone users in rich media?
Tier the data
center – do we need that many data centers all acting like tier 1, super cool,
super available, load-balanced? Can we not have the data center running at the
relevant temperature for the application or availability, if it’s Tier 2 or
Tier 3, (by that we mean world ending but not brand affecting or catastrophic),
can we not have those data centers slightly hotter on the basis that it might
save me millions of pounds for availability that isn’t needed, that by having
tier 3 running at 30 degrees, I might save a few million pounds a year in power
and cooling but have a marginal effect on availability, and any support cost
offset by the power saving. In this case for example, I could have data center
7 (which is 9-5 only) be powered down to low availability on minimal servers at
the weekend, and then bring all nodes online on Monday at 7 am in a controlled
fashion.
Virtualize the
application – abstract it to its component parts in workloads, data feeds and
user inputs, so that we can move it around the relevant load-balanced platforms
working on a shared infrastructure basis
Virtualize the
infrastructure towards shared infrastructure models where I buy the workload,
the capacity availability, performance or reliability I need, I only pay on
use, on the basis of application availability and reliability.
Virtualize the
storage and set standards to offline more data as it becomes less needed
online, by that I mean, we need to keep say 30 days trade data on the server
disks, the trade data going back three years is wonderful in terms of online
ness, but is an inefficient use of power and storage. At the same time, this
means we need a backup and recovery process that:
Works
Is on time
Is scalable and
enables recovery in hours not those, “Michael’s at lunch, we’ve ordered the
tapes, sometime next Thursday”
With a working
backup, I could offline more data to cheaper and more energy-efficient storage,
it might simply mean tape, it might mean cheaper disk backups for your last 6
months data, everything else on tape, etc – more efficient storage on a
per-application basis
Work with deduplication of data – how much space is taken up on the shared storage with user profiles, with static application data or copies of Office or other applications for user access, it might be more efficient operationally, but is this again because it takes too long to rebuild a pc? Without limits on user profiles, we could be copying gigabytes of history, temporary files, and user data around the network which might get backed up several times along the way.