Data centers (DCs) are rapidly transitioning from performance HDDs (hard disk drives) to SSDs (solid state drives), but understanding how to find the right capacity targets is critical in making the best choices. Accordingly, best practices for system sizing in many DCs are in need of a refresh. And, in some cases, DCs need a complete storage overhaul. As a result, running out of space or power can be a tremendous concern, and in many cases, quite an expensive problem if the most appropriate storage strategy is not put in place.
Of those DCs that employ 15K RPM HDDs, there is a distinct likelihood that most if not all are wasting OpX (Operational Expense), as large numbers of hard drives are needed to achieve ambitious performance targets. Conversely, significantly fewer SSDs are needed to accomplish the same goals.
Let me provide you with a concrete example, where we are sizing a rather typical storage system, one capable of a 20,000 IOPs peak workload with 100TB raw capacity.
HDD Sizing: As a starting point, calculate how many HDDs are needed to meet the performance target (20,000). In too many instances, you would typically select HDDs, each of which can handle 200 IOPs. Therefore 1,000 HDDs would be needed to handle the 20,000 IOPs.
Of course, your storage solution may not be comprised only of HDDs. Some DCs use storage arrays, but storage arrays are sophisticated configurations that usually are quite expensive. Many use caching to improve system performance. But caching typically only boosts performance when the data is already sitting there. When your data is not in the cache, there is no performance boost. For a deeper dive about this, here is a great resource.
As many of you know, 15K RPM HDDs generate 300-200 IOPs (outer diameter to inner diameter).
[The HDD data that I am using can be found here]
One thousand 300GB HDDs have a total capacity of 300TB. Yes, that’s 1,000 hard drives you have to buy to meet the performance target of 20K IOPs (mainly because of the low performance per HDD). As a result, you are having to buy 300TB of capacity, when you only really need 100TB.
SSD Sizing: For a starting point here, let’s calculate how many SSDs are needed to meet the capacity point. Then, we will confirm that overall performance significantly exceeds the target minimums.
If we size for capacity using 4TB SSDs, we would need 75 SSDs. But, we are buying exactly the capacity we need, and nothing more. Each SSD delivers around 100K/25K R/W IOPs (and I am using conservative numbers), so 75 SSDs will handle up to 7500K/1800K R/W IOPs. This exceeds our 20K IOP system target by more than 100x. Remember, that’s only 75 SSDs. As you can see, performance sizing of HDD-based systems results in over-buying on capacity.
When sizing SSD storage systems, however, you size for capacity first. When you do that, you will quickly see that you are buying only enough SSDs to meet your capacity target, while the delivered performance actually exceeds your target by a sizeable margin.
Price/Performance Winner: The advantage for SSDs gets even better when you expand the example analysis to the parameters of your complete storage system including additional servers, rack space, floor space, and greater power/cooling requirements. An attractive total cost of ownership, while still fulfilling business unit requirements, is a constant pursuit of every CIO.
Our example clearly shows why SSDs today are widely used from a performance perspective but are also the price/performance winners. I urge you not to delay in making your own comparative “total system“ storage analysis.