RAID system is an effective means to protect stored data. During RAID creation, system initialization takes a long time. Why does such an operation exist during RAID initialization? What impact does this have on SSDS?

All disks added to the RAID Group are divided into a series of slices based on their LBA addresses. These slices are called Stripe Units. Stripe units corresponding to the same LBA address on different disks are organized into a Stripe. Encoding all data in one strip, such as RAID6 producing two encoded data blocks P and Q, allows both data disks to be corrupted at the same time.

Therefore, in a RAID system, all data in a strip must meet the codec algorithm rules. That is, all data in a strip can generate encoded data according to certain rules, and the encoded data is the same as that stored in the strip. This is said to be the case where the data in the strip is consistent. When a disk fails, the lost data block can be recovered using encoded data stored in the strip.

If the data in a strip is inconsistent, that is, the encoded data in the strip is different from the encoded data stored in the strip. If a disk fails, the encoded data stored in the strip cannot be used to recover the lost data blocks. Therefore, a stripe with inconsistent data will cause data correctness problems when a fault occurs.

When you create a RAID system, the disks in the RAID Group may be new disks or used data disks. The data on these disks cannot contain all zeros. In this case, the data strips constructed by these disks cannot meet the requirements of data consistency. That is, the encoded data calculated according to certain rules in each strip is inconsistent with the encoded data in the strip.Inconsistent data striping poses great risks to RAID data correctness.

For this reason, when creating a RAID array, initialize all stripes in the system to ensure data consistency. Strip initialization can usually be done in two ways:

1. Initialize all strips in the RAID system by writing all zeros. The parity data of a strip with all zero data is also zero. Therefore, all zero data can ensure the consistency of the strip.

2. Perform parity calculation for all stripes and update parity data in the stripes to ensure data consistency.

RAID system initialization is a very long process. The main reason is that all the strips in the system need to be initialized. You also need to consider the performance balance with front-end user IO. Therefore, RAID system initialization is usually a background process that lasts a long time and affects the performance of front-end applications.

For SSDS, the RAID system initialization process also introduces other problems. During system initialization, data needs to be written to SSDS, no matter in zero write or parity data update mode. This process results in unnecessary data write enlargement. Before user data is written, a data mapping table is established inside the SSD through initialization. The service life and performance of SSDS are reduced. Therefore, a RAID system for SSDS needs to be optimized for the system initialization process, which is a special feature that traditional RAID does not take into account. Therefore, traditional RAID arrays cannot be directly deployed on SSDS, which affects SSD service life and performance.

RAID systems use striping to protect data, but a series of problems are also introduced in the process of striping data protection. System initialization is a typical problem of strip consistency. A good RAID Data protection system will solve this problem during the design process. For example, EMC Data Domain RAID does not have the system initialization process, of course, it needs to cooperate with the file system, and has done a lot of optimization in RAID strip data distribution.