Principles and implementation options for tiered storage architectures

Tiered storage is the allocation of different categories of data to different types of storage media to increase storage efficiency and reduce total cost of use (TCO). Storage classifications are basically based on the application's service level requirements, including availability, performance, retention requirements, usage frequency, and other factors. Tiered storage can be very complex due to a large number of growing electronic storage data, best practices policies, automated sustainability distribution software, specific data matching, and device features.

Tiered storage can take many forms and is often a natural consequence of the growth of storage architectures. By assigning the cache to different data, and/or by using a physically separate storage array of different characteristics, tiered storage can be built within an array (using different capacity or different performance disk drives).

Tiered storage, which claims to save up to 50% of the cost of storage, makes it an attractive option for undifferentiated capacity allocation. The main business drivers for tiered storage and end-user considerations include the following:

"SAN (storage LAN) heterogeneity" leads to incompatible islands of storage, and there is no good way to share data between servers and disk arrays;

Mergers and acquisitions have resulted in heterogeneous San architectures, further adding to complexity;

Expensive first-tier (TIER-1) architectures appear to be too costly at some times, enabling many companies to develop "avoid tier one" strategies;

Migration and configuration complexity requires applications to shut down in order to migrate data or configure new capacity, which often costs more than 50000 dollars per migration array.

Wikibon.org's end-user research shows that each dollar spent on hardware and software costs 50 cents for configuration during data migration and array usage. By tiered storage, this number can be reduced to $10 per dollar of hardware software spent on migrations and configurations.

In the mainframe world, automated tiered storage has been implemented for years, why is it so forgetful of its goals? The answer, of course, is the storage heterogeneity outside the host. Because of the fragmentation of storage hardware, applications, technologies, and architectures, several vendors are facing significant challenges in their efforts to automate. As a result, the market penetration rate of tiered storage is only 10% to 15% as a positive strategy, although, according to some people, virtually every company has a certain form of tiered storage.

One possible strategy is to migrate disk storage to a single schema (for example, all to tier one storage), but the cost of this approach is very high and most users cannot. The data format is just one of the challenges (for example, block vs file), not even the hardest problem. The problem that really hinders the adoption of tiered management is how to create an effective and "automated" policy-based, hierarchical classification system that is driven by the data access requirements of each application and user group.

To support this, some companies reduce the number of tiers they need to manage, communicate these guidelines clearly, and virtualize both the previous server resources and back-end storage assets, simplifying the strategy. One of the key benefits of virtualization is that applications can maintain the storage resources they see, but this perception can actually be mapped at any time-and dynamically. Data can be migrated seamlessly, and applications will not notice this change. However, virtualization also presents its own set of problems, including its implementation complexity, and the performance problems of many applications (discussed below).

For users who do not consider virtualization, their approach to tiered storage is to focus on the large data pools of homogeneous data-such as e-mail systems and software development data-and to build tiered storage islands around these rapidly growing data pools. This approach can also save a lot of costs, and its implementation is much easier than virtualization, but the cost of continuing management can grow very quickly. The balance approach is to manually build bridges for these data pools, and because applications need to know about changes, these applications need to be interrupted.

Virtualization is starting to build these bridges in an automated manner, and looks likely to be the best solution for many large enterprises in the future. However, as a scenario, several vendors provide a "built-in" tiered technology that allows higher-capacity, lower-cost devices to exist as a higher cost, higher performance, and lower-capacity drive in the same array. While this is one of the simplest forms of tiering, users are sometimes reluctant to take this approach, especially when it needs to add capacity to a more expensive tier-one storage platform.

It is important that while most of the tiered storage discussions focus on High-cost, High-performance tier-one storage, and lower-cost two-tier storage, the intermediate solution, more than half of the world's data exists in tier three systems, either very low-cost disks or tape technologies. This is a big problem for users, and must be taken into account in tiered storage policies such as records management and retention policies, which increasingly go into the day-to-day processes of storage managers.

Do what

Many companies interested in addressing San issues, such as slow, storage growth exceeding control, and unsustainable migration costs, are turning to tiered storage and taking the following measures:

Clearly define recovery point objectives (RPO) and recovery time objectives (RTO), and use this as a basis for data placement (a level of service that is relative to business-oriented and without clear objectives);

Communicate these requirements with the business unit to allow the IT department to allocate storage according to these policy guidelines;

The simplification level requires that the most stringent applications be placed at one level (depending on the service level), others by default at level two, and migrate to level three storage based on records management and retention policies, as required by law and company requirements;

Virtualize front-end and back-end resources, provide a service layer outside of the storage array, and gradually rely on low-cost arrays to reduce hardware costs and reduce expensive storage software license expenses. Virtualize all possible level two storage and maximize virtualization of tier one storage;

Storage management software and processes to reduce existing storage management software kits, if possible, to a suite;

Group By application, carefully test the reliability and performance of virtualization and deploy within a reasonable timeframe;