【Daily Question 20090404】Discuss the issue of server hard drive arrays.

RAID (Redundant Array of Inexpensive Disks) is a technology that uses one or more disk controllers to manage the interconnection of multiple hard drives, synchronizing their read and write operations in order to reduce errors and improve efficiency and reliability. RAID technology includes several standards such as RAID 0 through RAID 5, each with its own focus. The common standards are as follows: RAID 0 – This RAID level combines two or more hard drives, with the user data being divided into multiple manageable units. These units are distributed across different drives in the RAID 0 array. It’s just like a freight company transporting goods: in the past, there was only one vehicle for delivering goods, but now there are more vehicles that handle the delivery separately, so efficiency increases significantly. However, RAID 0 arrays do not store any redundant information; this means that if one of the hard drives fails, all data will be lost. Therefore, servers with high security requirements generally do not use RAID 0. RAID 1: In a RAID 1 system, the same data is stored on two hard drives (100% redundancy). When a disk drive fails, data can be immediately obtained from another disk, thus preserving data integrity. Just like when we print documents, to ensure that the data is not lost, we usually print an extra copy for storage. Additionally, when the raw data is busy, the system can also read data from backups; therefore, RAID 1 can improve read performance. It can be said that RAID1 improves both efficiency and the security of the system. RAID 2 distributes data in chunks across different hard drives, with each chunk being a bit or byte, and it uses a specialized encoding technique to provide error checking and recovery. RAID 2 is complex to implement, which is why it is rarely used these days; therefore, no further explanation will be given. RAID 3 is very similar to RAID 2, as in both cases data is divided and distributed across different hard drives. The difference lies in the fact that RAID 3 uses simple parity checking, with a separate disk used to store the parity information. If one of the data disks fails, the parity disk and its data disks can recalculate the lost data and restore it to the disk ； RAID 3 can provide a stable transfer rate for large amounts of sequential data, but for random data, the parity disk becomes a bottleneck for write operations. RAID 4 RAID 4 is very similar to RAID 0, with data divided among the disks. The difference is that RAID 4 uses one hard drive as a parity disk; when one of the data drives fails, the lost data is retrieved by calculating it using the remaining valid data drives and the parity information. This method of backing up data is similar to making a list of the files that have been printed, while printing them; in case one of those files is lost, it can be restored using the records in the list. This method is somewhat more complex than the backup approach used in RAID 1. A drawback of this technology is that every write operation requires access to the parity disk, which means that the parity disk becomes a bottleneck for write operations; as a result, RAID 4 is rarely used in commercial environments. The read and write processes for RAID 5 are similar to those of RAID 4; each write operation requires access to the parity disk, but in a RAID 5 array, the parity data is distributed across various disks. In this way, when there are multiple read and write operations simultaneously, each operation is distributed evenly across different disks, thereby providing a more balanced throughput. RAID 5 and RAID 4 have the same level of security: when one disk fails, all data remains fully accessible. The lost data is calculated from the valid data and parity check information. RAID 0 and RAID 1 are mainly used in home computers such as PCs due to their lower implementation costs ； RAID 2 is rarely used due to its complex implementation technology, so it is seldom employed these days ； RAID 3 and RAID 4 are suitable for large servers and video systems ； RAID 5 is primarily used in enterprises involved in large-scale data processing, such as financial institutions. Other types such as RAID 6, RAID 7, and RAID 10 were developed by various manufacturers, and there is no unified standard for them. The modes supported by the array chips or array cards that come standard with motherboards that we commonly use are: RAID 0, RAID 1, and RAID 0+1. As for what brands are available for RAID, I’m not sure, because it can be assembled by oneself. The RAID cards do have brands.

1. The commonly used hard drives today fall into several categories such as IDE, SATA, and SCSI; the main difference among them lies in their interfaces. The latter two, SATA and SCSI, are generally used on servers; there are also DIY setups that use the newer SAS hard drives. 2 1) RAID 0 uses disk segmentation to write data to multiple disks instead of just one disk; this is what is known as RAID 0. In a disk array subsystem, data is written to multiple disks sequentially, in units called “segments” as specified by the system. For example, Segment 1 is written to Disk 1, Segment 2 to Disk 2, Segment 3 to Disk 3, and so on. When the data has been written to the last disk, it starts writing again from the next available sector on Disk 1; this process of writing data is repeated until all the data has been written ； In simple terms, RAID 0 uses parallel access. 2) RAID 1: RAID 1 is known as disk mirroring – it involves mirroring the data on one disk to another. This approach ensures maximum reliability and recoverability of the system without affecting performance, and it offers high data redundancy. However, the disk utilization is only 50%, and the cost is high; it is therefore used mainly for storing critical and important data. However, in RAID as well, it consumes some CPU time to synchronize the data written to the main drive with that written to the mirror drive. 3) RAID 1+0 (also known as RAID 10): Although RAID 1 can also provide a slight performance improvement, it is likely minimal compared to RAID 0. Therefore, RAID 1+0 was developed to balance performance and security. RAID 1+0 uses at least 4 disks; thereby, it theoretically offers both the performance of RAID 0 and the reliability of RAID 1, at the cost of requiring twice as many disks as RAID 0 or RAID 1. 3. IBM, Samsung, Sony

RAID 0 offers the lowest level of security but the highest performance; unfortunately, it is hardly ever used in industrial servers. There’s no other reason; it’s simply because of its lowest level of security. RAID1 is the most common scenario and also the easiest RAID level to implement. RAID1 does not necessarily require SCSI or SAS. Currently, RAID1 based on SATA is just as useful, simple, and inexpensive. Many high-end industrial computers, DELL workstations, HP workstations, as well as DELL’s high-end business machines support SATA-based RAID1. Of course, many entry-level servers also use SATA-based RAID1. However, in most cases, this type of RAID1 is implemented using the southbridge chip on the motherboard; we see chips such as ICH7R, ICH8R, and ICH9R, with the ‘R’ indicating support for disk RAID. Often, this type of RAID is referred to by professionals as soft RAID. Think about it: a 640G SATA hard drive costs only 480 yuan, and that price keeps dropping. You only need to pay 2×480 to get 640G of storage capacity, thus providing double protection for your data. The cost is very low. It’s also easy to implement. There is no need to insert any additional RAID cards into the computer. At present, the biggest difference is the relatively poor performance; however, the difference in read and write performance at the RAID1 level isn’t as significant as people might think. Very small. The 2X1 RAID1 configuration allows for 1 failed drive. RIAD 0+1 is actually rarely used these days as well. Onboard RAID chips almost never support 0+1. Independent RAID cards and RAID array enclosures, on the other hand, tend to use RAID5 and RAID5+1 without exception. The aerospace and military industries prefer RAID6. RAID2, 3, and 4 are hardly used these days. Generally speaking, most current servers support RAID5. Each write operation in RAID5 requires a large amount of verification calculations. So RAID5 is always implemented in hardware. That is what professionals call a hard RAID. All of this work is handled by dedicated RAID cards. It is no longer involved by the CPU. The capacity calculation for RAID5 is (N-1)XM. Assuming there are 3 300G SAS hard drives, the resulting capacity will be (3-1) X 300G = 600G. Allowing one disk to be damaged. And only one disk is allowed to be damaged. For servers to use RAID5, at least 3 disks are generally required as a minimum. RAID5 cannot be created with fewer than 3 disks. However, RAID5 rarely uses more than 6 disks. Theoretically, it can go up to 8 disks. In fact, very few have more than 6. It’s mainly a matter of security and performance. With each read and write, after verification, the data is written to 6 disks; the load is not very high, but only 1 of those 6 disks is allowed to fail. The pressure is still quite high. The typical RAID5 configuration I’ve seen starts with 3 drives. 4 or 5 is very common. I have never seen one with 6 disks to date. The X3610 I have has 3 drives, while the X3650 has 4 drives. For industrial HMI servers, 300-500G is usually sufficient; therefore, RAID5 typically involves 3-5 disks of 300G each. But for those with a large amount of data to store. Generally, an external large-capacity disk array should be used. For example, Honeywell’s PHD, ABB’s IMS. KM. As well as systems such as OSI’s PI. And some file servers. Video file server. I have seen and heard that it is configured with 12 SAS 300G hard drives. Group, oh RAID5+1. 6 hard drives are used for RAID5, and another 6 are also in RAID5; the two sub-arrays are mirrored using RAID1. The available space is (6-1) X 300G = 1.5T. It should be noted that 12 disks were used, but only the capacity of 5 disks was obtained. This is the sacrifice made for the security of the data. Additionally, external RAID arrays have another significant use. Many multi-port RAID arrays can be used to connect cluster servers. A model with 2 ports can be connected to 2 clusters, while one with 4 ports can be connected to 4 clusters. There are no 8-port ones, ^_^. If connecting an 8-server cluster, it is common to connect the array’s 10GB high-bandwidth dual-redundancy interfaces to two switch-like devices, and then these two switches distribute the connections to the 8 servers.

I am still checking the information regarding RAID6; I’m not entirely sure about it at the moment. All I know for sure is that it applies when there are 4 hard drives. The safety factor of RAID6 is higher than that of RAID 0+1. With RAID6, when using 4 disks, the available capacity is still equivalent to that of 2 disks. However, RAID 0+1 allows up to 2 disks to fail, and only failed disks on different sides are permitted; failed disks on the same side are not allowed. In RAID6, out of the 4 disks, up to any 2 can be defective. The RAID brand? Haha, the simplest one is the southbridge chip integrated in Intel motherboards. Generally, mainstream server manufacturers have corresponding mainstream RAID vendor partners. For example, IBM prefers LSI Corporation’s products. Currently, the mainstream option is a 10IL RAID card. Its external array cabinet also has LSI controllers inside. Hard drive,,, originally an IBM hard drive, now a Hitachi hard drive; it also uses an LSI controller chip. DELL tends to opt for Adaptec’s OEM. It seems that all Perc series are OEMs from this company. There are many external array cabinets, and various independent RAID system manufacturers produce them, such as Promise. Highport, Adapterc. HP’s server RAID system is also OEM, though I haven’t figured out which company it is. For engineers who work in automation control. Many times, it is sufficient to know server brands with a good reputation. As long as you choose the right server, the RAID system provided by the server manufacturer generally won’t fail, haha. As long as communication is done in advance and the security level for the data is planned ahead, it is relatively easy to carry out.