Green Active Primary Storage

Posted by George Crump on July 1, 2009

How do you make primary active storage green? If you followed the advice of archiving old data as we discussed in Archiving Basics and have moved all that old data from primary storage to a less expensive tier of storage, you should be left with a much smaller but very active data set. These drives are and should be spinning, no MAID here or at least it has limited use because these systems should not spin down. So how do you green active primary storage?

For mechanical storage there are limited options that you can take. On this very active section of primary storage, capacity reduction has to be deployed carefully. It can not affect overall performance. One first step is to deploy storage systems that can utilize thin provisioning like those from NetApp, 3PAR and HDS, also for file systems consider deploying real time compression appliances like those from Storwize. In both cases these technologies can reduce the amount of primary storage and physical drives that are needed.

Reducing drive count really is the focal point for much of greening primary storage. Drives consume power, obviously, and the less you need to deploy the less power the storage system consumes. After thin provisioning and compression you can look at either increasing the capacity of those drives or replacing drives with Solid State Disk (SSD). In both cases you have to be cognizant of costs and performance impact.

As we explain in a recent video for Information Week one of the decision points is queue depth, basically how many outstanding requests does your storage have from applications? While we often speak of queue depth as a reason to go to SSD, it can be a reason to move to larger capacity drives as well. If you have a modestly performing application and you can still service its I/O requests with fewer but higher capacity drives, you end up with a big savings not only in power but in acquisition costs.

In most active primary storage cases however, the opposite is true, there are potentially hundreds of drives servicing the applications to eliminate or reduce the queue depth. At that same time a goal may be to lower response time, the amount of time it takes the drives to service each request. Once you are using high speed drives (15K) the only way to further lower response time is to short stroke those drives so that only the fastest third of the drive is used. Hundreds of high speed drives, formatted to 1/3 of their capacity all running full tilt is neither cheap nor green and this is where SSD comes in.