Why RAID Arrays Are Sinking

Sales of big RAID storage arrays boxes are declining quarter to quarter and this trend may be accelerating. It's worthwhile looking at the root causes of their sales decline and the long-term impact on storage buying strategies that this drop implies.

There are multiple forces at work hurting RAID array sales. Storage sales generally are weak, despite all of the hype around unstructured data the last couple of years. Most pundits point to macro-economic factors at work, such as the weak Chinese economy and the troubled state of Europe, and these clearly are having an impact.

There are, however, major changes in the IT industry that are driving systemic alterations in what storage is bought and how it is used. A number of technologies are factors, including solid-state drives and all-flash arrays, data deduplication/compression, larger capacity hard-disk drives, a new data redundancy model, hyperconverged systems, and the use of containers. Add in the impact of public clouds and the pressure on RAID revenue is enormous.

The flash storage factor

SSDs and all-flash arrays have changed the tiering rules in the system. Instead of a large primary RAID array tier of hard drives, we now have a relatively small, but extremely fast solid-state solution. This restructuring of the tiers recognizes that we only have a small amount of hot data in very fast storage and the rest can reside in bulk secondary storage. Contrast that with hard drive array configurations where the primary storage was perhaps 2x the speed of bulk drives. As a result, with hard drive arrays, the tiering model often was just a primary tier with cold data going off to disk-tape archiving.

Implementation of the new tiering is easy. Just plug an all-flash array into the SAN and the high-performance all-flash array, with its relatively small capacity, gives the whole server farm a new lease of life at a low price. The old HDD RAID arrays now are relegated to secondary storage, and in an ironic twist, the speed of the AFA left bandwidth for compressing and de-duplicating data stored in both  the AFA and secondary storage. 

This boosts effective capacity as much as 5x raw capacity, which lowers the acquisition cost of the AFA. It also boosts secondary storage enough to mitigate the need to buy more storage for a year or two in many cases, and certainly reduces the size of any new RAID array buys.

Upgrading secondary storage arrays to 4 TB or larger hard drives also reduces purchases. Admittedly, drives are still priced and delivered by the RAID vendor, but enterprises can avoid the cost of new arrays for capacity increases at least for a year or two.

All of these trends serve to reduce RAID demand, but RAID arrays themselves are doomed by the move towards SSDs. The bottom line is that a RAID array with 60 or 120 drives falls over from lack of bandwidth using just four or eight SSDs. Both network and controller bandwidths are way too low.

New data protection models

The result is we see many  storage appliances with a 12-drive configuration in a 2U box. These are essentially standard COTS servers, but the match of bandwidth to SSDs is much better. The driving force for the smaller appliance, however, is the change in data integrity models in the industry. RAID repair after a drive is lost has become a serious issue with traditional RAID approaches, and new models of data protection have appeared.

Replication and erasure coding both stripe data over multiple appliances, protecting against controller failure and allowing much faster, more parallel repair to occur. This lends itself to simple COTS-based single-controller appliances using inexpensive SATA drives. With very popular open-source code, these boxes are absorbing most of the object storage market, which is where organizations tend to store unstructured data .

From a revenue perspective, these are inexpensive boxes, supplied by many vendors, so the unit cost is low compared with a RAID array, as is the price per terabyte. Replacing large numbers of RAID arrays with inexpensive boxes makes a major dent in RAID and overall storage revenues.

Hyperconverged systems

Hyperconverged storage does away with the need for any fast RAID type storage altogether. The drives in hyperconverged appliances are shared across a virtual SAN instead. This means the only place for more traditional network storage  is in the bulk tier, where typically it is Ethernet connected and could be object storage or filer-based.

Containers remove the need to keep large images of the OS and app stack for each virtual machine, simplifying and reducing storage tremendously. In an extreme case, a VDI application may see a shrink of as much as 100x in total storage needed. This fits well with hyperconverged solutions, where the local storage can easily cope with both the size of the single image and the burst performance that VDI demands. Local instance storage in a virtual cluster can achieve the same result in non-container environments.

The cloud is the great specter at the feast for RAID vendors. Cloud providers roll their own architectures, while the dominance of REST-based object storage in the cloud blocks most use of any commercially available RAID arrays for cloud storage.

The decline of traditional RAID arrays is systemic. The advent of much larger and cheaper bulk SSDs in 2017, coupled with the ascendency of Ethernet over Fibre Channel for storage, will accelerate that decline.