Thin Provisioning as discussed in Storage Switzerland’s recent article on Thin Provisioning Basics, is the ability to define volume sizes in advance but to only allocate that capacity dynamically. Doing so reduces the up front storage requirements of an application and can dramatically reduce the amount of storage that needs to be purchased. An aspect of the subject that is glossed over is converting from fat volumes to thin provisioning. How can the old volumes be converted to take advantage of the new thin provisioning technology?

Over the last year thin provisioning has quickly become a minimum bar of entry for customers selecting a new storage system. Not only does a thin provisioned system reduce the amount of storage required upfront it also, when properly implemented, can greatly reduce the amount of time storage managers spend provisioning and managing storage.

Properly implemented is the key term. In the race to catch up with market innovators like 3PAR, some companies are gluing together components of their solutions to make it look like they have thin provisioning. As we pointed out in the Thin Provisioning Basics article, it should allocate very small blocks of storage as new writes occur, and should not require pools with pre-dedicated capacity. The companies with true thin provisioning capabilities are now enhancing the capability with the introduction of new technology and establishing a Thin Provisioning 2.0 standard.

Thin Provisioning 1.0 focused primarily on new volumes with new data leveraging “dedicate-on-write” technology.  Migrating existing volumes to thin aware volumes, however, was challenging and represents a high-return initiative with the right technology. Thin Provisioning 2.0 and specifically “zero-detection” technology holds the key to delivering the ability to convert existing volumes to thin volumes and rid data centers of hundreds of terabytes of unneeded capacity.

Converting from Fat to Thin

Thin volumes have been ideal for getting new projects underway with a minimum of investment. Of course most customers that were considering a new storage system would also want to migrate existing data off of older arrays to the new storage system over time. The typical and most practical SAN to SAN migration strategy is using a tool that does a block-by-block copy from the old array to the new one. These tools offer a much higher performance data migration option versus file copy approaches.

There are two issues that these utilities must contend with. First and foremost is when the volume is created by a storage system that is not thin, i.e. a hard volume. Essentially the unused capacity within the volume is filled with zero’s awaiting the eventual possibility of an application writing data to it. The problem with general migration utilities is that when a block-by-block copy is conducted, it will propagate the capacity inefficiencies by writing zero-blocks to the new storage, even though it is unused space.  Even if the array supports thin provisioning, a data migration of a traditional or hard volume writes to every block associated with the volume, negating the benefits of thin provisioning. One or two utilities can now handle zero’ed out blocks of information and will not write them during a migration, but they require servers be on the latest software version and are limited to a couple of operating systems.

Until recently, the only alternative to migrating a fat volume to a thin volume is to backup the old fat volume at a file level, and then to restore the files to the new thin volume. Since the backup job is file based, the empty blocks and blocks containing deleted information within the fat volume would not be carried over to the new thin volume. The challenge with this migration method, even with disk backup, is of course time. This migration is a long slow process that imposes disruption to the application.

There is another way—by leveraging a Thin 2.0 technology called zero-detection, organizations can achieve a host-independent Fat-to-Thin conversion. Zero detection technology can be software-based or, ideally, hardware-based.  The drawback of zero detection in software is that the XOR calculations done by the array to find and “virtualize” unused capacity can consume large amounts of controller performance.  Fat to thin conversions could therefore take a very long time and affect performance of other volumes within the array. 

A hardware-based approach offers several advantages, an example of which is 3PAR whereby a dedicated “zero detection” processor has been incorporated into its array controllers to find and virtualize unused capacity at wire speeds.  The most obvious benefit of moving the zero detection technology to a specific hardware ASIC is performance, not just of the conversion itself but it allows the rest of the system to focus on storage IO for its other volumes.  In addition the silicon-based solution can work independently of various operating systems and migration software versions. 

Thin provisioning is ready for an upgrade. Thin 2.0 will deliver greater utilization of the storage asset while delivering better performance yet requiring less people to manage. The reality of the new economy now makes this more critical than ever. Converting from Fat Volumes To Thin Provisioning is the first aspect of Thin Provisioning 2.0. Our next article, we will take a look at yet another innovation around thin provisioning on the way.  Is Thin 3.0 on the horizon?