Tiered storage is the process of assigning progressively less-expensive storage categories to progressively less-valuable data. It’s up to IT to classify storage tiers using a matrix of performance, price (Capex and Opex), storage capacity and data services. Classifying data priority is not entirely up to IT. Within the same storage system, automated tiering functions will classify data by features like I/O patterns and move it accordingly within the storage system’s internal storage tiers.
However, IT will need to assign data priority by business need in order to migrate data effectively throughout the storage infrastructure, ultimately landing in highly cost-effective cold storage. Different companies will assign different data priorities according to their business and compliance needs.
Aging is the most common metadata for demoting data to less expensive storage, but other factors may affect the outcome. For example, IT may progressively demote aging data and eventually add it to cold storage on tape or cloud. But some aging data may reside long-term on SATA on-premises storage because it is subject to regular information audits.
An effective overall tiering process largely depends on automating data movement across the storage infrastructure. If IT is spending a lot of time manually tiering data — or not tiering at all because it takes too much time— then the company will not see value out of storage tiering. Automated storage tiering across data lifecycles is a real necessity for getting the kind of value that IT needs out of long-term data storage.
The Difference between Storage Tiering and Flash Caching
Let’s talk about a perennial point of confusion: the difference between tiering and caching. When you’re talking about tiering data through the entire storage infrastructure, there is no question that you’re talking about storage tiering. But when you’re talking about dynamic data movement on a single storage system, then the difference between tiering and caching is not as obvious.
The big difference between the two is that caching copies or mirrors frequently accessed data to flash drives, whereas tiering moves data. There are, of course, distinctions between caching products: some are block and some are file-based; some operate on the server side and some from the storage system; some have only a single flash tier and others contain internal SSDs for storing cached data. But in general, cache products copy data to a high-performance flash tier to accelerate applications.
Dynamic storage tiering within the array does not copy data, but moves it between storage tiers based on access patterns. If all you are after is I/O performance in a single system, then the caching model may work well for you (or you should consider buying an AFA). However, if you want both accelerate performance and realize more cost-effective storage usage, tiering will go farther and accomplish more. Unlike a caching layer that keeps flash media otherwise unusable, tiering will deliver data to faster flash tiers for high-performance processing, but unlike a cache tier the storage is usable and not just a copy of the data. Then as data access patterns change, the tiering software will move it to less-expensive storage tiers for cheaper retention.
Storage Tiering Domains
Vendors offer automated tiering in very different locations and configurations. The most common choices include within the hypervisor, within a single storage system or connected same-vendor systems, or across the entire storage infrastructure. Most companies will invest in all three approaches depending on their needs.
In the Hypervisor
Hypervisor-based storage tiering optimizes virtual machine (VM) performance by dynamically placing VMs in differing storage classes. High-performance VMs reside on high-performance storage tiers, while those that don’t require lowest latency reside on less-expensive tiers. VMware Storage Dynamic Resource Scheduler (Storage DRS) typically manages live tiering across the VMware environment. DRS recommends placement based on I/O loads and storage constraints in order to reduce latency impact. Default placement is every eight hours but admins can customize the schedule and assign storage classes.
Microsoft uses features in Windows Server 2012 R2 and Hyper-V Server 2012 R2 to automate storage tiering. This process continuously analyzes I/O and moves blocks accordingly between flash and hard disk tiers.