Flash is fallible. But you'd rather have an AFA than spinning rust

A touch of the SSD magnetic

Latency wins the day

For many purposes, flash's low latency wins the day. For instance, it means you need fewer "drives" to achieve the same performance. It is very common in enterprise storage to have to gang dozen of disks together, because each spindle can only deliver a certain number of megabytes per second.

As disks grow in size this means lots of wasted space, as disks are added for their incremental bandwidth, not their data capacity. Substituting flash enables entire racks of spinning disks to be replaced with far smaller boxes, with all the power and space saving that implies. Storage developers are already seeing flash displace high-end 15K RPM hard disks, as it does a better job for much the same cost.

Low latency also allows you to apply data de-duplication and compression techniques to primary storage, despite their computational overhead. This helps all-flash storage arrays to become price-competitive with disk arrays, because it enables 1GB of flash to do the work of 4GB of disk or more. (The one caveat here is if your application must operate within a regulatory framework that requires data to be stored uncompressed, in the belief that compressed data is no longer verifiably original.)

And most importantly for many applications, flash means consistent quality of service (QoS). Flash can serve random IO far better than disk can, and in a multi-tasking environment – which these days means pretty much anywhere, but especially enterprise and virtualised applications – that means no waiting while the storage serves another application.

This is particularly important if you have noisy neighbours, which means applications that try to seize all the available IO bandwidth, probably because they were written with a single-tasking environment in mind.

One other point to note is that a lot now depends on the flash memory controller, or SSD processor. Indeed, the controller is in many ways more important than the type of flash, whether it is lower density, high endurance SLC (single level cell), higher density but less reliable MLC (enterprise multi-level cell), or slower but high density TLC (triple level cell). It is the controller that enables the use of more dense flash by applying error correction, and it is the quality of the controller's algorithms that determines data density, minimises write amplification and so on.

The best of both worlds

With the right storage architecture behind it, it doesn't take a lot of flash to provide almost all of the available benefits. In many applications therefore the most cost-effective approach is a hybrid one – flash at the front for its speed and low latency, and something else at the back for cheap bulk storage. Hot data resides on flash, then once it is no longer being accessed or updated, cold data is moved onto longer term storage, typically arrays of SATA hard disks.

Flash can also be implemented in several places within the overall infrastructure. While a tier zero of flash within the storage arrays is the most obvious, you can also have server-side flash, using PCIe SSD cards to accelerate applications even before they hit the network.

This can be particularly effective if your application and its data fits within the server SSD. And you can have flash in the storage network, acting as a network cache; the advantage here is that it is independent of both the servers and the storage arrays, and should work with your existing systems.

Longer term, look for flash to gradually move into more and more of your storage tiers. Early adopters are already using four tiers of flash in different configurations, and high availability all-flash systems are winning acceptance among cloud providers, thanks to their reliability, high density, QoS and cost-effective scalability, serving multiple customers using fewer drives.

Lastly, don't forget that it is not only production systems that can benefit from flash. It can also profoundly change your speed of development and your agility, and therefore your go-to-market speed. Development and deployment in a flash, now there's a motto to run with!