SSDs: Understand The Options

Windows 8 Preview: Key Features

I'm going to be in Denver in a couple of weeks speaking to a group of end users on how to deal with the various storage challenges they are facing. Of the three topics, one that is getting a lot of attention is our chalk talk on solid state devices (SSD). There is an overwhelming number of options for the data center to consider. Our goal is to cut through the marketing and to make sure you know how to balance your performance improvement with your spending. Here are a couple of hot topics we'll be addressing, if you attend:

-- I/O operations per second (IOPS) are overrated. Rule number one: don't focus on IOPS. I've grown tired of vendors claiming x number of IOPS for their storage systems. The reality is that the overwhelming number of environments can't come anywhere close to achieving the reported numbers. The workloads that vendors use to generate those IOPS are not the typical ones you see in the data center.

Reducing latency and in some cases increasing bandwidth are the two metrics that will more than likely improve your performance. A very simple example is upgrading a laptop from a hard drive to an SSD. You probably noticed faster boots and application loads. But this was driven by reduced latency, not increased IOPS. Laptops simply don't have the number of parallel storage requests required to drive IOPs.

[ For another wrinkle, read All-Flash Systems Vs. SSD Appliances. ]

The same is true with most servers--yes, even virtual hosts. The performance improvement you will get from SSD is most often a direct result of reduced latency, not IOPS.

Instead it might be better to measure what the SSD device does do for you. How much faster does it make your application or server host? This can be seen in reduced time to execute certain queries or long batch report jobs. The SSD also can be measured by how much more of something can you do. For example, a host might be able to comfortably support a dozen busy virtual machines with standard mechanical hard drives so the need for SSD might not be apparent. But if an SSD-based solution allows that same host to support three to four times the number of virtual machines then it will be less expensive to add an SSD to the environment. Basically, fewer hosts with SSD--instead of increasing the host population.

-- Fewer hosts. Designing an infrastructure that supports a very dense--100:1--virtual machine-to-host ratio is another top topic in our live sessions that we'll cover in Denver, too. As I noted in my recent article Increasing VM Density with All-Flash Storage, flash-based storage systems can dramatically increase the number of virtual machines per host and save money by reducing the number of hosts. However, the design requires a more holistic approach than just adding shared flash storage.

The number of options in the SSD market can be overwhelming. There are options for SSD in the server, network and storage system, and knowing which one or group of these solutions to implement is a daunting task. The hype around IOPS makes the selection process even more difficult. Focusing on what matters--making your applications more responsive or your environment more scalable--is an important first step.

Big data places heavy demands on storage infrastructure. In the new, all-digital Big Storage issue of InformationWeek Government, find out how federal agencies must adapt their architectures and policies to optimize it all. Also, we explain why tape storage continues to survive and thrive.