Triangulating speed, cost and reliability

Warren Reid explains what customers want and why they often do not get it

If you ask the channel what customers want at the moment, it's all about SSD. The high-speed dazzling performance of flash is a big attraction to any firm wanting higher performance in the datacentre. But there are several reasons this won't be the option that many companies end up with – or at least not the only option.

Primarily, it does not satisfy all three criteria in a golden triangle: speed and cost as well as reliability.

The cost of SSD is still prohibitive for many companies. In a networked RAID system, speed can be increased through the choice of drive type.

These range from cheap 7,000rpm HDD (7k SAS costing about 10p per Gigabyte) to 10k/15k drives at about 45p per Gigabyte, and at the top end, SSD costing perhaps 10 times as much as HDD, at $6.18 (£3.78) per Gigabyte for the equivalent capacities.

However, SSD is far more cost effective in performance terms.

Each customer has different priorities for storage. For some, speed is crucial because it affects productivity – for example, in video post-production, where the time taken to process and render video footage is significant.

My rule is that the storage infrastructure should be able to cope with the peak traffic workload while still providing acceptable response times for all applications.

It makes little sense for most companies to store all their data on SSD. In practice, the most used and critical applications can go on flash, with the rest on other storage media.

With all-flash networked RAID arrays, customers need to consider how much of their data really needs SSD levels of performance. Although the all-flash array will provide very low latency, there is a point quite early on where adding more SSD will not make the data throughput any faster.

This is because the SSD performance will quickly saturate the bandwidth of the RAID controllers. So for many customers a hybrid approach, combining both SSD and HDD with intelligent real-time auto-tiering, is gaining popularity.

An alternative to the all-flash networked array is to include flash technology within the server, either on a PCIe card or with the emerging NVDIMM form factor, which offers even lower latency.

This approach can certainly give incredible performance and is the solution of choice for many companies that want excellent database performance.

However, moving primary storage back into the server may represent a single point of failure. With the server-based flash approach, cost can really escalate if you attempt to strengthen all three sides of the golden triangle.

Many customers choose the cheapest RAID system they can find, on the theory that RAID protection will save them from data loss. However, in the event of a disaster, the low-cost and less-proven RAID software may fail to deliver the goods.

RAID is data protection but not backup – which means the possibility of losing data should the RAID controllers fail is high.

There is a lot of debate right now about the efficiency of RAID as a data protection mechanism, certainly as drive capacities continue to lengthen rebuild times in the event of a RAID-protected drive failure.

Applications that require massive capacity may be scaled out, using new ways to distribute redundancy to maintain performance and reliability for massive installations.

For most users the traditional scale-up style of deployment will prevail, with a continual flow of enterprise class data protection and operational convenience features finding their way into entry-level storage products.

The suitability of an all-HDD storage array for general purposes has reached the end of the line for most business-critical applications. SSD is not set to replace HDD any time soon but the technologies will co-exist, used intelligently, for some years.

More often than not, customers compromise and adopt a tiered approach to managing their data, in the most cost-effective way on the most suitable disk drives for the data. That is, SSD for primary data and HDD or tape for data less accessed but nevertheless needed.

This is just as well, as according to analysts there are simply not enough SSD production facilities should all major companies decide to migrate their existing storage to flash.

And so the golden triangle of speed, cost and reliability tends to be delivered with a combined, compromise media approach. The job for the channel in 2014 will continue to be to help find the best golden triangle outcome for each customer.

Warren Reid is EMEA marketing director at Dot Hill Systems