During NetApp’s Q3 fiscal year 2010 earnings call in February, CEO Tom Georgens said “the entire concept of tiering [storage] is dying.” He went on to say, “with…our performance acceleration module, basically systems are going to go to a large amount of flash which are going to be dynamic with serial ATA behind them and the whole concept of having tiered storage is going to go away.”
These remarks sparked a fury of response that swarmed the Internet and a number of bloggers, reporters, analysts and industry observers have debated over caching versus data tiering as the ideal method for managing and moving data to the optimal storage.
To Tier or Not to Tier
In an ideal world where a CIO does not have any budget, space or energy consumption limitations, companies would not need to tier storage because an array would be filled with solid state drives (SSD) for optimal performance, and there would be no need for Fibre Channel (FC), SAS or SATA drives.
In reality, analysts estimate 80 percent of all data is inactive, meaning that a large portion of data being stored has not been accessed in the past 30 days. Therefore, most of an organization’s data needlessly sits on high-performance, expensive storage, such as SSD or 15,000 RPM FC or SAS drives.
The challenge IT managers face is how to deliver high performance for demanding applications without wasting disk space. Some vendors are promoting the caching of data to flash memory as a solution.
All storage arrays first store data in cache as a buffer before writing the data onto disk drives. Some storage vendors use cache acceleration I/O cards to extract the most active data from the system memory and hold that data in a second buffer that is read-only. To improve performance for read-intensive applications such as e-mail, the system first checks whether the requested data has been cached in one of its installed modules. Without this special cache, if the data is not in system memory the array next tries to access the disk drives, which adds latency.
While this caching approach can boost I/O performance, the architecture can be difficult to expand and manage over time as a company’s data grows. As memory technology evolves, to increase the size of the read-only cache, users are faced with a choice of discarding or replacing old modules to take advantage of newer technology. To add to the complexity, the high-capacity solid state modules are typically only supported in the newest systems, preventing customers with older arrays from accessing the most recent technology unless they perform a rip and replace upgrade.
Given some of these limitations, automated tiering becomes a more viable solution for managing and moving data.
Automated Tiered Storage
Tiered storage moves the most active data onto expensive, fast disks and the inactive data onto cheaper, slower disks with higher capacity.
However, not all storage systems tier in the same manner. Some products may require manual intervention or move data at the volume level — meaning entire applications such as e-mail have to be dedicated to a single tier, which can be highly wasteful in terms of both dollars and disk space. By automating the data movement at a granular level inside the volume, companies can achieve the right balance of performance, scalability, cost and ease of management benefits.
For example, in a Compellent Fluid Data system, detailed information about each block is captured in action, providing system intelligence inside the volume such as the last time the block was accessed, the type of drive used, and the RAID level.
Compellent automatically classifies and migrates the data to the optimum tier of storage based on actual usage. Active blocks of data are written to Tier 1 storage such as SSD, or 15,000 RPM SAS or FC drives, with performance-optimized RAID levels such as RAID 10. Inactive data is automatically migrated to lower-tier storage with higher-capacity and high-protection RAID 5 or 6.
When tiering is done automatically and with minimal management from the IT personnel, the benefits of keeping active data on high-performance disks can be realized. Users can eliminate manual data classification and migration while lowering drive counts, as well as cutting power and cooling costs. They can mix SSD, FC, SAS and SATA drives in the same system, and SAS drives with multiple rotational speeds in the same enclosure. For mission critical applications, users can maximize write performance and data availability with solid state storage and add disk drives to any tier of storage on the fly.
The Future is Built in
Ultimately, drive technology will continue to evolve, but even as it does, demands on storage will continue to grow as well. As the capacities of current high-performance drives increase, other higher performing, expensive and lower-density drives will be introduced to the market. In an automated tiered storage solution, customers can mix and match the storage drive that meets their needs without requiring an expensive and time-consuming upgrade every time a new technology is introduced. On the other hand, the cache-centered approach locks users into a cycle of adding and throwing away modules whenever the cache is maxed out or if they want to migrate to a different storage technology.
In the end, tiering is not a dying concept, it is a fundamental one. It will be around until the IT department does not have any budget, space or energy constraints to consider. Being able to automatically tier their data across multiple disk classes and RAID levels allows companies to tune their storage to meet application demands. As a result, IT can automatically shift resources where they are needed, accelerating the speed of change to support the ever-increasing pace of business.
www.compellent.com
 ShareThis
Tags: Tiered Storage, Disk/RAID/Tape/SSDs |
