Solid State Drives (SSD) for Enterprise Storage – A new vocabulary is required

This is the first in a series of blog entries on how SSD’s are requiring those of us in I.T. to understand that SSD drives operate unlike traditional mechanical hard drives. To make informed purchasing decisions, you will need to understand this new vocabulary which describes the important characteristics of SSD drives. I hope to enlighten you so that when your peers ( or vendors ) start waving shiny new SSD drives in front of you, an informed decision can be made.
Most of us in the I.T. industry have a pretty good handle on understanding our drive choices when choosing traditional mechanical hard drives. SAS as a drive interface is a given, in most cases. We then may choose inexpensive 7200 rpm drives for slower archival storage, or 10,000 rpm drives for our mainstream storage. High performance needs may dictate the use of the most expensive 15,000 rpm drives. Other than budget considerations, this pretty much wraps up the decisions we need to make with traditional hard drives. Life was so easy back then.

Well, here is our first problem, SSD drives don’t spin. We all know they use solid state memory, so they probably perform fairly similarly, right? Well actually no they don’t, so today’s topic is about the i/o performance differences between SSD drives. Our word for today, which may not be new to you, is IOPS ( Input/Output Operations per Second ). Simply put, IOPS are measured by using a tool, like IOMeter, to read and write data to a drive to measure its performance. With traditional drives, the results were very predictable, with faster spinning drives giving you fairly linear improvements in performance. Results such as “150 IOPs” ( Eye-Ops ) were pretty typical, with no brand or model departing dramatically from these numbers.

So let us look at the HP “HP Enterprise Solid State Drives” Data Sheet, Rev 5, from February 2015, to get some current IOPs numbers on SSD drives. Please note the IOPs ranges vary depending on the specific drive model ( part number ).

HP 6G SATA Value Endurance Hot Plug SFF (2.5”) SSD drives

• Random Write IOPs from 8000 to 35,000
• Random Read IOPs from 58,000 to 64,000

HP 12G SAS Mainstream Endurance Hot Plug SFF (2.5”) SSD drives

• Random Write IOPs from 42,000 to 64,000
• Random Read IOPs from 66,000 to 87,000

HP 12G SAS High Endurance Hot Plug SFF (2.5”) SSD drives

• Random Write IOPs from 66,000 to 73,000
• Random Read IOPs from 84,000 to 87,000

So clearly as we move from the family of “Value”, to “Mainstream”, to “Endurance” drives, performance increases. As there is such a wide variation in performance even within each family of drives, we may choose a drive not only on its family, and data capacity, but also the performance of the specific drive. Most dramatically these results show that Write performance varies from 8000 IOPs at the bottom end, to 73,000 IOPs at the top end, a performance difference of almost ten times! This is important, as Writes to disk are one of the slowest operations traditionally, yet it may commonly represent half of our disk i/o.

You might look at these numbers and think that since even the slowest Enterprise SSD is 50x faster than your hard drives, it is a win-win to go with the least expensive SSDs. Sorry, that isn’t the correct conclusion, as we have more SSD vocabulary to talk about. Next time we learn about an even more important difference between how an SSD drive functions compared to a mechanical hard drive, and why IOPs are likely not the most important factor to consider.

In a later blog, I’m going to demonstrate even further that in a real world, where Enterprise SSD’s are expensive, that there is another reason why more IOPs, and more dollars, may not produce a significantly better user experience.