A NEW BREED

storageage
Scaling out of the past
The ability for a storage set-up to scale according to the data at an
organisation’s disposal is a vital component of modern infrastructure
I
n today’s big data age,
organisations long for a
storage system that can
support data growth at a
scalable and manageable cost.
Cloud computing has presented one
answer, but the security and privacy
concerns around trusting a third
party with business-critical data
has left many organisations seeking
an alternative.
Scale-out storage has emerged as a
popular option. As the name suggests,
this solution allows organisations to
expand their total amount of disk space
as required, delivering predicted and
manageable levels of performance,
capacity and cost.
The result, vendors claim, is a flexible
and dynamic storage environment that
can support balanced data growth
and reconfigure infrastructure on an
as-needed basis.
Such a set-up is underpinned by the
ability of multiple storage arrays to
connect over Ethernet. While a
traditional unit is autonomous and
cannot share or receive data from other
arrays, the connectivity between units
in a scale-out structure means users
have multiple boxes that can work
together as a clustered storage array.
‘The faster the communications link,
the more performant the cluster
will be,’ says John Abel, engineered
systems leader for the UK, Ireland and
Israel at Oracle. ‘Scale-out storage is
‘Scale-out storage
is ideally suited to
handling linear growth
in performance
needs, which is where
traditional storage risks
falling into bottlenecks’
>> John Abel, Oracle
ideally suited to handling linear
growth in performance needs, which
is where traditional storage risks
falling into bottlenecks.
‘Of course, as you continue to add
arrays, you are also taking on more
physical storage, which demands more
overhead and management costs.’
The key technologies that make up
scale-out storage are the management
software that orchestrates the nodes in
the cluster and the network that links
the nodes together.
Both are designed to deliver coherent
data across the nodes and ensure that
in the event of reliability issues data
remains accessible and consistent as
nodes drop out of the cluster.
The design point that various
products solve differently – and with
different levels of success – relates to
the CAP theorem.
‘This states broadly that you can’t
have consistency, availability and
partitioning tolerance at any given
time, just two out of the three,’ says
Alex McMullan, EMEA chief technology
officer at Pure Storage. ‘That drives
multiple copies of data for resilience
across nodes, increasing cluster traffic
and, ultimately, cost.’
Old and new
Traditional storage infrastructure is
often based on proprietary hardware
that was never designed to scale
intuitively to meet fluctuating demand
for capacity.
Often developed to be deployed from
a single location, most organisations’
legacy infrastructure is designed to
manage a consistent flow of basic data.
This makes it rigid to varying
demands and, thus, likely to choke
when expected to cope with vast
amounts of data – like that produced as
a consequence of big data initiatives or
Internet of Things (IoT) deployments.
September 15 information-age.com 27