Automated Tiering Methods

A few entries ago we opened up the subject of Automated Tiering with an explanation of why the technology is becoming so needed. As this series of entries continues we will review various storage vendors specific approach to automated tiering, but first it is helpful to understand the common methods that are employed.Automated tiering is the dynamic placement of data on different classes of storage based on parameters that define how that data is being used, most often its level of activity that defines where the data goes. These classes of storage typically range from some form of memory based storage (RAM, FLASH SSD or DRAM SSD) to Fibre/SAS mechanical drives to SATA drives. Depending on the vendor they either supply all of these classes of storage or they merely provide the automated tiering intelligence and you provide the classes of storage.

Most of the focus with automated tiering is moving active data to the fastest tier possible. The idea is maximize the benefits of the most expensive and fastest class of storage. If you are paying more per GB for the memory based tier of storage then you want to make sure you buy as little as you have to and that it is almost always near full. Running memory based storage at 50% utilization is a significant waste of resources.

The first method of automated tiering is to treat this faster tier of storage as a large cache, similar to cache technologies that already exist on drives and storage systems today. The main difference is that they are significantly larger. The concept has merit. Cache technology is certainly well vetted yet vendors can still add value by customizing the approach. It can have a safer feel to it as well by using it in a read only mode, meaning that if the automated tiering device fails you have not lost data. Of course that also means in write heavy environments you would see no performance benefits.

Most of the caching systems and all of the second method of automated tiering solutions have the ability to treat this higher speed tier as something more permanent. Data will reside uniquely on a particular tier for a significant time. That time could be a few seconds, in the case of cache based systems, up to a few days on the second method, which I'll call the storage method for lack of a better term. The storage method systems also typically have a tunable setting that allows you to set how long data is uniquely on each tier of storage. While this method should lead to further performance boosts it may also lead to data loss if the automated tiering device fails or the tier which has the data fails. Typically though the storage method systems provide for some HA (highly available) functionality.

Next up we will look at the different protocols that are supported (file and block) as well as the level of granularity (block, file, LUN) that these solutions tend to offer.

Track us on Twitter: http://twitter.com/storageswiss

Subscribe to our RSS feed.

George Crump is lead analyst of Storage Switzerland, an IT analyst firm focused on the storage and virtualization segments. Find Storage Switzerland's disclosure statement here.