The Capacity Creep Dilemma
Posted by Damien Bowersock on Thu, Dec 17, 2009 @ 07:05 AM
Apparently, judging by the response to my previous post,
everyone is struggling to find ways to reduce the storage explosion in their
data centers. So, let's discuss
one of the major problems in data storage that has a fairly simple solution. A problem I call the "Capacity Creep
Dilemma."
The Capacity Creep Dilemma occurs when the following
scenario plays out in an IT organization:
1.
A developer, tasked with creating a new
application, requests storage for his/her new project. In this example, will use a 100GB
storage requirement.
2.
The storage administrator, receiving this
request, thinks to himself, "I know this person is always wrong so I'll setup
200GB." And, having no available capacity on the SAN, submits an expansion
request to purchase a new tray of disks.
After running his RAID calculator for RAID5 with 146GB drives and a hot
spare, he nets out a request for 4 drives or 600GB.
3.
The storage manager, having dealt with the
storage administrator before, knows that there are probably other requests that
haven't been fulfilled and it's cheaper to buy a full shelf of 15 drives than a
partially populated one. Therefore,
requests from purchasing the acquisition of 15x146GB drives or 2190GB of
storage.
4.
The purchasing, privy to the year-end "deal"
gets a buy-one-get-one offer and submits an order for two shelves, which yields
nearly 4.5TB of storage.
This case may seem extreme, but these scenarios play out
every day. Take a serious look at
you current SAN and calculate the amount of data stored for the primary copy
versus the physical disk it resides on.
Odds are your real utilization rate is around 25-30%. And that would be the best-case
scenario.
So, how do you solve this problem? The reality is, unless you are willing to make fundamental
changes in architecture and habit; you may not. But the good news is technology has changed. Most of these rituals are predicated on
10-12 years of SAN purchasing habits.
But, the technology that we have today is not the same Fibre Channel
based architectures of yesteryear.
It is time to move from physically tying the data to the
drive. By creating an abstraction
layer between the physical device and data set, customers can provision volumes
more quickly and increase utilizations.
One of the big problems in the past was getting the spindle count high
enough to hit performance needs, but not have too much excess capacity lying
around. By utilizing a high-speed
file system such as GBFS you can have the best of both worlds. GreenBytes allows you to have both NAS
and block-level storage presented without physically tying you down to an
individual spindle or spindle set.
Everyone knows the actuator is the slowest moving part - why strap your
performance to it?
And to think, we haven't even touched on improving efficiency
with Thin Provisioning or Deduplication.
Take all the functionality that virtual servers have brought to the
server team and apply it to storage.
With GreenBytes, you can.