RAID 1
Wed, 10/17/2007 - 08:51 — Tomas Hlatky
RAID 1 - mirrored set; data are duplicated on two or more different drives. This set up is for redundancy to keep your data effectively backed up at all times, however you are loosing the possibility of using of maximal drives capacity. Big advantage is, that if either drive fails, the other continue to function (with lower performance). Probability of the failure of all combined drives is multiplied by the probability of the failure of each single disk: pRAID1 = p1 × p2 × ... × pn, where n = number of drives. For example, if you have two drives set up in RAID 1, each drive has p = 0,001, probability of all the setup failure is 0,0012 = 10-6. However, my advice is don't always rely on theory, and also don't be absolutely sure that redundancy will protect your data completely. It can happen, that identical series of disks can have the same production error; in this case it's likely that, all disks can fail in one time. There are also outer conditions, which can influence RAID 1 function, as overload or blackout.
Array capacity has size of smaller drive.
The output depends upon correct implementation - in optimal case, sequential read performance is multiplied by number of disks combined in the array, as different parts of one file are read from different drives in one time. However, it might happen, that due to older or low-end controller, rebuilding can be slower. What is more, you have to skip in between blocks with file fragments, what can cause lower sequential read performance. There is a slight degradation of write performance comparing to single drive write performance, as total body of data is multiplied due to the writing on each single drive combined in the array and this process can slower the overall output. There are few administrative advantages of RAID 1 trim. For example, in several systems, there is a possibility to disconnect one drive (marked as disactivated) backup and replace it. After replacing, it's necessary to rebuild the array. This might be useful for systems, where is needed to have files always available (to keep the consistency of data, it is not possible to write on the disk, during backup)
2 x ST3500320NS & RAID 1
There are two ST3500320NS disks set up in RAID 1 on the first screenshot. Unfortunately, ICH8R chipset is not able to read data from both drives, and thus speed up a data reading. Like this, in HDTachRW, two drives combined in RAID 1, show very similar performance as one disk, burst speed is only 151,1 MB/s. On the second screenshot, you can follow two ST3500320NS drives, combined in RAID 1, with W-BC on. Recorded results in HDTachRW are just too bad, only burst speed is 1792,2 MB/s, what is not so bad, but still comparing to 2060,4 MB/s recorded for RAID 0, it is markedly lower. Let's have a look what two ES.2 disk set up in RAID 1, are going to do in application tests.
Unfortunately, also according to application tests, two disks set up in RAID 1 are somewhat (10%) slower that single disk, and this is even more remarkable when WB-C function is on. Users requiring high data security have to count with this downside (at least with Intel RAID controller). I highly don't recommend starting the WB-C function. I failed, while setting up three disks in RAID 1 from Intel(r) Matrix Storage Console.
Array capacity has size of smaller drive.
The output depends upon correct implementation - in optimal case, sequential read performance is multiplied by number of disks combined in the array, as different parts of one file are read from different drives in one time. However, it might happen, that due to older or low-end controller, rebuilding can be slower. What is more, you have to skip in between blocks with file fragments, what can cause lower sequential read performance. There is a slight degradation of write performance comparing to single drive write performance, as total body of data is multiplied due to the writing on each single drive combined in the array and this process can slower the overall output. There are few administrative advantages of RAID 1 trim. For example, in several systems, there is a possibility to disconnect one drive (marked as disactivated) backup and replace it. After replacing, it's necessary to rebuild the array. This might be useful for systems, where is needed to have files always available (to keep the consistency of data, it is not possible to write on the disk, during backup)
- Advantages: total data redundancy; increased read transfer rate
- Disadvantages: no increase in volume capacity; slight decrease in write transfer rate
- Applications: typically used in workstations and servers to store critical data
2 x ST3500320NS & RAID 1
Create RAID Volume Wizard
2 x ST3500320NS in RAID 1
2 x ST3500320NS in RAID 1 & Enabled Volume Write-Back Cache
There are two ST3500320NS disks set up in RAID 1 on the first screenshot. Unfortunately, ICH8R chipset is not able to read data from both drives, and thus speed up a data reading. Like this, in HDTachRW, two drives combined in RAID 1, show very similar performance as one disk, burst speed is only 151,1 MB/s. On the second screenshot, you can follow two ST3500320NS drives, combined in RAID 1, with W-BC on. Recorded results in HDTachRW are just too bad, only burst speed is 1792,2 MB/s, what is not so bad, but still comparing to 2060,4 MB/s recorded for RAID 0, it is markedly lower. Let's have a look what two ES.2 disk set up in RAID 1, are going to do in application tests.
Unfortunately, also according to application tests, two disks set up in RAID 1 are somewhat (10%) slower that single disk, and this is even more remarkable when WB-C function is on. Users requiring high data security have to count with this downside (at least with Intel RAID controller). I highly don't recommend starting the WB-C function. I failed, while setting up three disks in RAID 1 from Intel(r) Matrix Storage Console.
