Shulou(Shulou.com)06/01 Report--

Architect Technical Alliance

We who deal with data should know that data in the storage system transmission, after a number of components, a variety of transmission channels and complex software processing process, any one of the link errors may lead to data errors. But this kind of error generally can not be detected immediately, but later found that the data has gone wrong through the application in the process of accessing the data. It is very difficult for this kind of data to be detected at the moment of the data error. We call it silent data corruption, that is, Silent Data Corruption.

Causes of silent data corruption

There are many reasons for silent data corruption, but they can be roughly classified into the following categories.

Hardware errors: memory, CPU, hard disk, data transfer link, etc.

Firmware error: HBA, hard disk, etc.

Software bug: system software, operating system, applications, etc.

Other factors: such as noise, electromagnetism and so on.

Data consistency standards and organization

In 2007, DII (Data Integrity Initiative) was founded by Emulex, Oracle, LSI and Seagate. DITWG (SNIA Data Integrity Working Group) is established by SNIA. They focus on two technologies, T10 Protection Information-DIF and Data Integrity Extensions-DIX.

The T10 standard protects data integrity by adding protection information (PI,Protection Information) to each data block, which was once called data integrity field (DIF,Data Integrity Field). 8 bytes of protection information is extended in each logical sector to ensure data consistency. 8 bytes include 2 bytes of Logical Block Guard,2 byte Logical Block Application Tag and 4 bytes of Logical Block Reference Tag.

T10 PI only includes data protection from the host HBA card to the hard disk through the storage array, DIX in order to extend the scope of DIF protection. Data integrity protection is extended to the application layer to HBA. DIX uses the same 8-byte data integrity information as the T10 PI as the data check field. The difference is that IP Checksum is used as the Logical Block Guard in DIX to reduce the computing overhead of the host CPU.

DIX+DIF can realize end-to-end data protection from application to hard disk. DIX ensures the data integrity of applications and HBA cards, and T10 PI (DIF) ensures the data integrity of HBA, arrays and hard drives.

DIX and DIF data read and write process

The additional 8-byte check data added to data integrity is divided into several segments, which is called DIF on the storage side, and later renamed T10 PI; on the host side as DIX. When writing data, the host HBA bus adapter, array target chip or other components generate 8-byte PI according to the user data. During the data transmission, the data will go through a checkpoint to check whether the data and PI match. If errors are found, errors will be returned up. If there are no errors, they will continue to be transferred down and finally written to the hard disk.

Write data flow: when the data is written to the host memory, Oracle ASM library will add 8 bytes of DIX check to every 512 bytes of data, and the 8-byte check will go through the OS along with the IO request to reach the HBA card driver; after the DIX check of the HBA card, delete the DIX check, and generate 8-byte PI checksum data to be sent to the array, the array verifies the data integrity, and sends the data to the hard disk.

Read data flow: read the data and T10 PI from the hard disk and verify the integrity. If errors are found, the data is rebuilt and repaired by RAID, and if there are no errors, the data continues to be transferred upwards. After T10 PI check, HBA deletes T10 PI and generates DIX protection information to return to the host. The DIX protection information goes through the OS and returns to the application layer along with the IO request. ASM Library verifies data and DIX protection information.

Necessary conditions for supporting DIX

To use the number consistency feature, first, the array needs to support standard T10 PI, even if it does not support DIX, or it can support T10 PI; on storage. Secondly, it requires specific configuration of database, operating system, HBA and other components.

Database: Oracle 11g or above.

OS:Oracle Linux 5 or 6 running the UEK2-200 kernel.

Specific models of FC HBA cards for HBA:Emulex and Qlogic

Storage vendors that support DIX

As DIX and P10 are widely used in enterprise Oracle databases and are a bright spot in storage, mainstream storage manufacturers support this feature, such as EMC VNX series supporting custom PI, VMAX supporting standard T10 PI, and supporting DIX;HDS HUS series supporting custom PI, HDS VSP supporting standard T10 PI,DS5000 supporting standard T10 PI,DS5000, some specific models supporting PI;HP P10000 supporting standard T10 PI, Huawei OceanStor 18000 and V3 full range.

Whether the file system needs DIF

It is worth mentioning that the T10 PI requires 520 sectors of the disk to support (512 for data storage and 8 bytes for T10 PI check data), which Seagate supports. The general file system does not need DIF, because the file system manages data through metadata, which will cause the page data layout of IO to change constantly, and the database and network card manufacturers are not strong enough to form similar specifications and standards. Search for "ICT_Architect" to join the Wechat official account "architect Technology Alliance" for more exciting content.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.