Cost Optimizing for Performance

There has been a lot of attention paid lately to using capacity optimization tools and techniques to optimize storage capacity. With the combination of 1 TB drives (now mainstream), 2 TB drives around the corner, and space optimization technologies like data deduplication, single instance storage, and inline data compression, users have a lot of options for holding all their archive and backup data for a very long time. This is good for the large percentage of data in the enterprise where standard performance is more than adequate, but what about data that does need performance? How do you cost optimize that?

Interestingly, the data where performance does matter is often relatively small when compared with the rest of the data set -- and, really, that is the problem. You almost can't get the capacity small enough, and the workarounds to generate better performance with today's technology are very expensive.

Short stroking drives is a good example. This is the technique of formatting just the outer platter of a drive, often a high RPM drive, to maximize the response time of the drive. The problem is that these drives are already expensive, and formatting them down to two-thirds their normal capacity makes them even more so, especially when you consider that it often takes 50 or more of these drives to generate the performance required by a high-end application.

Wide striping is another technique that some of the virtualized storage systems like 3PAR and Compellent use. This gives you the ability to build high drive count array groups without having to sacrifice the capacity as you do with short stroking. Performance is cost optimized by leveraging the remaining capacity of these drives for the standard performance data set. It doesn't hurt that even that data is likley to see a performance boost as well. Wide striping is an important capability to consider, especially if you are looking to upgrade your primary storage system and want to either delay or negate the need to implement solid state disks (SSDs).

As discussed in our Byte and Switch article on knowing when to jump to SSDs, some environments are going to reach a point where adding more drives won't increase performance at all, and increasing parallelism by adding servers will be too expensive. In these situations, SSDs not only will often eliminate the performance bottleneck, they will also actually cost less than the short stroked alternative and will have an added side benefit of reducing power and cooling costs.Like any emerging technology, SSDs have been "emerging" for a long time, and there are different ways to implement them. There are the specialized systems from suppliers like Texas Memory and Solid Access; there is the EMC and, soon, many others' method of adding drives to an existing storage solution; and there are the PCI-E attached systems by companies like Violin and Fusion-io.

Stay tuned for a deeper dive into each of these methods.

To learn more about Solid State Disks, please view our Storage Switzerland Webcast.

George Crump is founder of Storage Switzerland , which provides strategic consulting and analysis to storage users, suppliers, and integrators. Prior to Storage Switzerland, he was CTO at one of the nation's largest integrators.