Don't start reading the last rites for monolithic storage just yet
But other tech can simply do the job better, sorry
Commercial off-the-shelf (COTS) storage shelves and servers are used together with dedicated software that does a better, and lower-cost, job of unstructured data handling and organisation than the monolithic array controller code stacks, and their supplier pricing structures, can provide.
Bulk nearline data is going to a variety of places but the commonest destination is to clustered dual-controller arrays with multiple data structures supported. These have been clustered for some time; think NetApp ONTAP clusters, some have more than a pair of controllers. All are growing in scale.
The ability to scale is prized by customers who don't want to buy a high-end system now, although they may well need its capacity in five year's time. They prefer to buy array capacity in increments of both performance and capacity, taking advantage of steadily lowering $/GB capacity costs, and the improving processor performance offered by Intel as it follows Moore's Law.
This they can do with a modular system, or a clustered dual-controller system, but generally not with a monolithic system.
Fujitsu's latest offering, the ETERNUS DX8900 S3, is an 2-24-controller, system that can house up to 16 petabytes of block-access data using more than 4,600 2.5-inch drives or SSDs. ONTAP clusters can also go as high as 24 nodes, but are limited to eight nodes for block-access storage. With the DX8900 supporting up to almost 14 petabytes of storage the capacity direction is firmly upwards.
File data stored in monolithic arrays is migrating to modular arrays and to specialist scale-out filers, such as the Isilon range, which are better able to handle large numbers of files by, for example, having the file:folder metadata stored in fast-access flash memory, and thus getting to the actual file data without enduring disk access latency delays.
Where there is less need for extreme scale-out file performance then software-only storage, the ZFS-based NexentaStor can be used. Our feeling is that this is often chosen as an alternative to costlier modular storage than monolithic storage.
Another alternative to both monolithic and modular storage arrays is to use hyper-converged systems. These combine servers, storage and networking in single systems, orderable with a single reference number (SKU) and operating and managed as single entities rather than three co-located boxes. The VMware-defined EVO: RAIL systems are examples of such systems as are alternatives from Nutanix and Simplivity.
Their claimed advantage is simplicity, with owners not having to separately buy, operate and maintain servers, storage and networking hardware and software.
They are cheaper to buy than the separate components and simpler to grow; just add another node to your cluster. Their software is newer than that of traditional arrays and can be more efficient in its use of, for example, SSDs.
The software and hardware can be extensively instrumented so that system operating data is collected and analysed by the supplier with developing fault situations detected and fixed before parts break.
But ... these systems are inflexible in that you don't have the freedom to mix and match best-of-breed components yourself and you are locked in to the overall hardware/software system supplier, or software supplier if a software-driven architectural template is used.
IT history indicates that such concerns occur and re-occur regularly and we can expect them to come around the block again when the first flush of enthusiasm for hyper-converged systems is over.
Another alternative to monolithic, and indeed on-premises storage, is storage in the cloud supporting applications running in the cloud. Amazon, Microsoft Azure and Google are the main suppliers, and they generally build their own storage using COTS hardware and software rather than buying monolithic or modular arrays.