Information Technology Reference
In-Depth Information
of the same infrastructure share computing resources with each other unknowingly because
of abstraction. Each user sees only the part of the infrastructure and resources allocated to
him, so it would appear to each one that they are the only users of the system.
So like all shared computing resources, shared storage is simply a large pool of storage
capacity that is collected from all connected devices within the infrastructure that provide
storage, such as host servers, SANs, and NAS.
Clustered Storage
Clustered storage, or grid storage, is a relatively new paradigm in storage technology
that pushes the efficiency and scalability of storage area networks (SANs) to new and
higher levels. It is based on concepts used in clustered computing, making use of large
storage farms that are linked in a grid fashion through horizontal scaling with more
storage servers rather than vertically using servers with more storage capacity.
Clustered storage systems are made up of smaller network-connected storage systems
with administrative functions that manage individual disks to act as a collective whole. To
a user client or a server making use of the clustered storage, the abstract collection appears
as a single large pool of storage capacity. There may be more control because clients can be
allowed to access smaller portions of the pool such as a single SAN or even micromanage
single servers or disks. These storage clusters can be made available to the whole network
for use, and each client is able to create a logical volume (like a virtual disk) on demand.
The beauty of clustered storage is the modularity and failure tolerance; you can add and
remove resources without affecting the whole pool. And because the smaller clusters that
make up the whole can be located in different geographical areas, outages can be isolated
to certain locations and clusters without adversely affecting the whole. Clustered storage
would be highly available and fault tolerant.
But common clustered storage offerings available in the market today are still using legacy
architectures. This older clustering technology requires a master controller that coordinates
tasks among the different nodes in the cluster. This introduces a possible bottleneck and a
single point of failure for the whole system, and it may also limit scalability because adding
more servers also means that more nodes are now contending with the master controller's
resources, which will eventually impede the performance of the whole system, a point where
diminishing returns occur.
The newer clustering technology removes the central master by essentially creating
multiple masters that control smaller portions of the cluster. These masters act as peers
that, when combined with a distributed file system, are able to work in parallel with each
other to enable sharing, selection, and aggregation of storage resources, which are likely
to be distributed in different administrative domains. This method allows us to horizon-
tally scale almost infinitely by simply adding new master controllers when more capacity
is added and eliminating the single point of failure.
Clustered storage is the best approach to delivering cloud computing benefits of scaling
as storage technology becomes cheaper and it becomes easier to add more resources into
the pool.
Search WWH ::
Custom Search