But even if that never happens, SSDs are revolutionizing data tiering anyway because they allow storage engineers to create two distinct classes of storage: a SATA tier lowering your cost of capacity ($/GB) and an SSD tier lowering your cost of performance ($/IOPS). No middle tier of storage is necessary.
By creating pools of storage and enabling data tiering, you can promote small chunks of active data to high performance disks and demote small chunks of inactive data to high capacity disks. From the chart below, for the same price (in this case, $150,000), you can either purchase 100 FC15 disks or purchase a combination of 80 SATA and 20 Solid State Disks. The latter configuration gives you 4.4x performance (IOPS) and 2.7x capacity for exactly the same price. That’s compelling!
NOTE3: The pricing in the example above (and in the chart below) is representative and is based on some industry best practices. The actual price is not nearly as important as the relative price between the different types of disks. Pay attention to how the price ratios change over the next couple of years!
More importantly, with tiering enabled, you can let the tiering mechanism decide which chunks of data should reside on which set of disks. Set it and forget it.
Of course, tiering has its limits. One major limitation is reaction time. How frequently do data chunks get moved? If it’s in real time like EMC Symmetrix FAST-VP, that potentially steals CPU cycles from the array. If so, you may be reluctant to enable tiering for your most demanding workloads (which would benefit the most from tiering). If it is, say, every 24 hours like IBM EasyTier, how useful is that to your users? With many of today’s popular applications, hot data chunks have very short and/or very unpredictable demand patterns. Think VDI or data analytics or email.
Several vendors are increasingly implementing longer term tiering schedules, especially in their midrange arrays. They are also enabling the storage engineer to choose the right chunk size for their environment. This combination works because vendors are combining tiering with another emerging trend: disk-based caching.
Tiering vs Caching
Not long ago, only the most expensive, high end, enterprise class storage arrays offered a generous amount of high speed cache to optimize I/O operations. Today, with the declining price point of SSDs, even most midrange arrays now offer a combination of a modest amount of very fast cache (usually PCIe-based) and the option to use SSDs to provide a generous amount of second tier (ie lower speed) cache before data even hits the storage disks.
Disk-based caching is fundamentally different from tiering. Tiering decides where to place *relatively* large data CHUNKS according to *relatively* long term access patterns. But data caching immediately places very small BLOCKS of *transient* data on designated SSDs as soon as that data has been requested more than once. Newly requested data blocks are constantly replacing older, less popular, data blocks on these special purpose cache disks.
As you can see, caching is a perfect fast, real time complement to tiering’s long term trending algorithms. But one leading storage vendor has proclaimed that tiering is a fad that is not necessary and has decided to double down on caching instead. Can you guess who that vendor is? Hint: they were one of the first to implement storage virtualization. We’ll discuss that in Part 2.