Thin provisioning

from Wikipedia, the free encyclopedia

Thin Provisioning (TP) is a cost-saving process for providing storage capacity in virtualized storage environments ( storage - virtualization ). In contrast to the usual full allocation ( thick provisioning ), only the memory that is actually required is reserved with lean memory allocation. As a result, it appears that more memory can be made available than is actually available.

Classic memory allocation

When administrators first start using a storage system, they must:

With this classic method, which is also called thick provisioning, hard provisioning, fat provisioning or exact provisioning, storage capacity is allocated to users and application programs . Another term, dedicate-on-allocation (reserve when allocating), describes the actual process.

Partitions and logical hard disks defined in this way can no longer be changed using operating system functions. If the allocated storage capacity is exhausted (this is usually the case at 80 percent utilization level, as additional space is required when reorganizing databases , for example ), all data must first be copied to another medium and then the LUN must be deleted and recreated. With special software tools, experienced users can also correct the size of a LUN including all data during operation. For security reasons, data backup should not be dispensed with.

While it is not a problem for private users to reserve the entire storage capacity of a hard drive for their needs, storage administrators in larger companies must be more careful with resources . You only allocate the storage space that is likely to be required to users over the life of a project. In order to avoid changes to these specifications and thus costly administrative work, the allocated capacities are generously dimensioned to be on the safe side. Part of the data growth is attributable to this fixed reservation. The amount of data actually stored is often only a fraction of the reserved capacity.

Modern memory allocation (dedication-on-write)

Thin provisioning can be compared to the electricity made available by an electricity company. This guarantees the customer maximum performance that he can obtain. Nevertheless, the E-Werk can only provide part of this maximum output to all customers at the same time. But since they never all draw maximum power at the same time, there are no problems.

In the IT sector, the term is often referred to as overcommitment or over-quoting.

Thin provisioning makes use of the fact that modern storage systems make virtual hard disks available. The customer (server) is provided with more capacity than is reserved in the storage system. As soon as the capacity used by the server exceeds a certain threshold, additional free capacity is made available for the customer from an existing storage pool.

There are basically two different ways of providing disk capacity:

Block-based storage provisioning or file-based storage provisioning. The SNIA has the Shared Storage Model developed a description of these different methods. The difference lies in the type of access.

With the block-based storage system , the hard disk is divided into logical blocks - these blocks are made available as a logical unit with an associated logical unit number ( LUN ). This can be done using different protocols such as B. Fiber Channel or iSCSI . The server (recipient) usually uses this LUN to store a file system such as. B. NTFS , ext3 , VxFS to write. In this case, thin provisioning is implemented by the storage system. The server sees a LUN that is x GB in size. In fact, the storage system has only reserved xn GB. If the server needs more capacity than xn, the LUN is automatically enlarged from a free pool. Unlike a file system, however, the LUN is expanded again with each additional write from the server - regardless of whether the server has meanwhile deleted data on the file system based on the LUN. This has to do with the fact that the server has no means of communication with the storage system that would allow it to inform the storage system that certain blocks can be released. Some memory manufacturers offer special software with which a reorganization can be carried out - however, this is only possible if no new writing takes place during the reorganization. Almost all storage systems evaluate the so-called SCSI TRIM command with which the file system driver can inform the layers below which blocks can be viewed as "free". Since 2015 there are no more operating systems that do not support this TRIM procedure.

With the file-based storage system , the hard disk is made available by a server or filer as a file system using protocols such as NFS or CIFS . The providing server in turn uses block-based LUN itself. Provisioning takes place here via the filer. Almost all manufacturers of file-based storage systems offer such functionality. In contrast to block-based thin provisioning, the filer itself is the master of the file system it provides. If data is deleted, he can independently release the capacities again. However, this has no influence on the effectiveness of file-based or block-based solutions since all operating systems are now able to use this method (see above).

The advantage is obvious: Less storage capacity is required - which has a direct impact on price, floor space, air conditioning requirements, power consumption, etc. The risk is that very detailed monitoring has to be implemented. As with the electricity company, new capacity must be provided quickly if necessary. The E-Werk achieves this, for example, by purchasing overseas in the event of overloads - physical disks have to be bought in the storage area. If usage increases faster than new capacity can be made available, the application comes to a standstill. In this respect, not only monitoring, but also trend line analysis is an important prerequisite for using thin provisioning.

Footnotes

  1. ↑ SNIA Shared Storage Model (PDF; 384 kB)
  2. ↑ Under IBM Redbook Thin Provisioning in an IBM SAN or IP SAN Environment (PDF; 600 kB) you will find a very good description of this effect and how quickly the positive effect of thin provisioning can be canceled out with normal disk usage.

literature

  • Kai Qian, Letian Yi, Jiwu Shu: ThinStore: Out-of-Band Virtualization with Thin Provisioning . In: 2011 Sixth IEEE International Conference on Networking, Architecture, and Storage . July 2011, p. 1-10 , doi : 10.1109 / NAS.2011.39 .