The challenge with some hybrid arrays is that they aren't appropriate for all workloads. For example, the do not deliver predictably high service levels and low latency. If data is on flash, then life is good, but if application data is sitting on an HDD tier, this can result in lag time for end users. Why? Because hybrid arrays use different algorithms for balancing data placement across flash and HDDs. How well these algorithms will do often isn't known until the array is used in production. Also, over time the efficiency and effectiveness of data placement may vary when application workloads change or more data is added to the storage array.
One recent variation that has been deployed for arrays that support flash caching is combining auto-tiering with flash as a cache. By adding flash as a cache, some of the challenges around auto-tiering can be mitigated. This way, data in a cache can be quickly accessed by the host even if it's in a cold tier.
When performance matters most, nothing can beat an all-flash array built on a flash-optimized architecture. All-flash storage can deliver consistent performance and, when combined with the right data services, do so without compromising resiliency, scalability, or data mobility.
And don't get fooled by the perceived higher initial costs of flash. As a report from IDC put it: "Solid state is a more expensive storage media compared with hard disk drives on a raw dollar-per-gigabyte basis ($/GB) but can be significantly more cost effective on a $/IOP basis. When solid state storage is intelligently integrated into a system with storage optimization technologies, storage vendors lower the acquisition cost and total cost of ownership (TCO)." This lower cost is referred to as the "usable dollar per gigabyte."
However, there can still be a performance bottleneck depending on the system design. This is why simply filling traditional dual-controller arrays designed for HDDs with flash drives is not the best idea. You will overwhelm the ability of the storage controllers to keep up with the SSDs, resulting in architectural bottlenecks that act as a performance throttle. Because of this, newer flash arrays may vary widely, not just when it comes to performance, but also in the consistency and predictability of that performance. Multi-controller, scale-out designs are in favor as they provide greater ability to keep up with performance demand.
Converged Flash Arrays
A newer concept in the flash world, the converged flash array is a category of all-flash array that can accommodate disk media as well. By utilizing a flash-first design and associated data efficiency technologies to reduce the cost of flash, organizations are putting complete applications and data volumes on flash and leaving them there. However, for secondary applications and older data, sometimes enterprises still desire HDDs within the same system. The ability to deliver both in a single system offers the performance and latency of an all-flash array, the affordability of a hybrid array, the agility of a unified array (with support for block workloads, file shares, and object access), and the scale and resiliency of a high-end array when the right data services are applied.
Which of these is right for you, of course, depends on your particular needs. With the above information, you should be better equipped to make the best decision for your business.
Sign up for CIO Asia eNewsletters.