Content Copyright © 2008 Bloor. All Rights Reserved.
For me, thin
provisioning of storage is a logical extension to virtualising it. Early on,
vendors who offered thin provisioning solutions had to struggle against
scepticism backed by FUD from some of the big hardware vendors not
over-enthusiastic about a technology which helped customers buy less tin.
happened. Gradually there was greater adoption of hardware-software solutions offering
thin provisioning until, this year, most of the leaders now offer some thin
provisioning flavour. IBM even offers it in its SAN Volume Controller (SVC) appliance—which effectively makes it available across multiple vendors’ back-end hardware
if used through an SVC.
Thin provisioning means
allocating disk storage space only as it is actually needed—which avoids purchasing
extra capacity only for it to sit idle (but spinning and using energy). Typically,
only 25% of the total physical disk allocation for a standard system is actually
needed. This makes thin provisioning a very green technology, reducing expansion
of disk resources (footprint), energy and heat output—and therefore cost. Equally,
since the provisioning can be largely automatic, it can greatly reduce the need
for management of disk resources and capacity planning.
So are these tough
economic times the point at which thin provisioning is really going to take off? Sadly,
I suspect not, at least in a big way. While the cost savings that may be made
once installed sit fairly well with the economic downturn, the capital investment
and a trade-off in performance may muddy the ROI case.
issue falls into a couple of categories. The first is that if you increase individual
disk utilisation because the data is spread across less platters you also increase
disk contention and so lower overall throughput performance. This may not
matter for most applications but does for the mission-critical ones,
particularly if there are SLAs associated with them.
Systems’ thin provisioning system counters this problem with a quality of
service (QoS) feature so that critical applications can be given message
queuing priority over others; this can be fine tuned. It is, of course, an
extra job for the storage manager to consider.
A second issue is
the type of application and storage being dealt with. Database management
systems tend to grab their whole allocated space straight away; even if there
is a work-around it could complicate things—or it may be a case of applying thin
provisioning only to part of the storage—which increases the management load
in deciding what can and cannot be included and perhaps copying a few files around
as well. Yet a major selling point of thin provisioning is that it can automatically deal with space allocation
issues so that the storage manager does not have to.
Another possible factor
is how well the software can utilise freed-up disk space following deletion of
old data. Total thin provisioning space savings are increased if the software
can handle this automatically, but some systems cannot. But, for instance, Compellent’s
thin provisioning feature eliminates allocated but unused capacity as it goes while
3PAR’s InServe architecture inherently reallocates unused blocks. (Having said
this, companies may not all delete
enough data to make this very fruitful.)
In addition, if
yours is an enterprise with mixed vendor equipment, you may think long and hard
before installing yet another vendor’s purpose-built thin provisioning hardware/software
solution, knowing that you will have to redirect all the files and databases to
be included to the new system. It might mean a fair amount of planning so that
realising a real return could take a while. Incumbents which can add thin provisioning to existing hardware may do better.
thin provisioning offering is different and most only work on their own specific
hardware. For the reasons given, it may also be difficult to clearly demonstrate
a fast ROI so vendors will need to work closely with their potential clients in
Thin provisioning has a chance of benefiting from the credit
crunch—some think it will be a fat one.