There are several different RAID levels available to servers (and even desktop PCs) with today's technology. These offer varying features to either enhance performance or reliability, or sometimes a combination of both.
Use the guide below to choose the RAID level that suits your needs. All quoted usable capacities are approximate.
| RAID Level | Performance | Reliability | Recommendation |
|---|---|---|---|
| 0 | Good | Poor | Not recommended |
| 1 | Improved read speeds | OK | Acceptable for non-demanding uses |
| 5 | Improved read and write | OK | Superseded by RAID 6 |
| 6 | Good, with modern controllers | Good | Suitable for most applications, can suffer from long rebuild times though |
| 10 | Excellent | OK | Good performance, good rebuild times but high cost. |
| 50 | Very good | OK | Capacity loss/expense, not used in most situations |
| 60 | Excellent | Good | Capacity loss/expense, not used in most situations |
| JBOD | Single drive performance | Average | Used with an upstream RAID controller |
RAID 0, also known as striping leads to improved performance as the workload is shared between two or more drives. However it has no fault tolerance. If any of the member disks develop a problem the RAID may fail and data corruption or loss is likely.
Example: Two 500GB drives in a RAID 0 gives a usable capacity of 1TB (1000GB).
The mirror has improved fault tolerance over RAID 1. Each drive is a duplicate copy. The system can benefit from improved read speeds with a controller which can read alternate blocks from each drive at the same time. Write performance is the same as a single drive, since write operations must be duplicated.
With RAID1 the capacity of the RAID is halved.
Example: Two 500GB drives in a RAID 1 gives a usable capacity of 500GB.
RAID 5 used to be the common RAID standard for servers. Write and read performance is improved with each additional disk or spindle that you add to the RAID. You lose the capacity of one drive, since one drive contains parity information. (Technically, the parity information is distributed across all of the drives, however you do loose one drives worth of overall capacity). You need a minimum of 3 hard drives for a RAID 5.
The disadvantage of RAID5 is that if you suffer an outright hard drive failure, the remaining drives need to be in perfect working order with no bad blocks. RAID6 gets around this problem.
RAID 5 is computationally expensive as parity must be calculated. Therefore it is slower than RAID 0.
Example: Three 500GB drives in a RAID 5 gives a usable capacity of 1TB (1000GB).
RAID 6 takes RAID 5 one level further with two lots of distributed parity. You need a minimum of four hard drives and will lose the capacity of two. RAID6 has even more computational requirements over RAID5 and for this we recommend a dedicated hardware RAID controller that has the offload capabilities.
RAID 6 can suffer two hard drive failures and preserve the data. Typically if you have one outright hard drive failure, you can still rebuild the RAID array even if bad blocks are encountered on other drives during the rebuild, as long as the bad blocks do not appear in the same location on multiple drives. RAID 6 has improved reliability against RAID5 and is recommended for all general application and file storage requirements, including VMWare datastores.
Example: Four 500GB drives in a RAID 6 gives a usable capacity of 1TB (1000GB).
Example: Five 500GB drives in a RAID 6 gives a usable capacity of 1.5TB (1500GB).
When a drive fails in a RAID6 array the entire RAID6 pack needs to be rebuilt. This involves reading the contents of all remaining working disks to write the contents of the newly fitted replacement disk. This can lead to degradation in overall system performance for an extended period of time whilst the rebuild is carried out.
In this situation, RAID 10 can give improved rebuild times and greater overall performance, but with the negative impact of increased cost or reduced capacity.
These RAID levels take smaller stripes (RAID 0) and then RAID these together either with RAID 0, RAID 5 or RAID 6.
Because of the increased (doubled) hard drive count, performance can be impressive. However high cost is a negative factor with these RAID levels. RAID50 can only withstand a fault in one stripe at a time so if you need an ultimate performance storage system we recommend RAID6.
Example: Six 500GB drives in a RAID 50 gives a usable capacity of 2TB (2000GB) (RAID 5 of three 1GB stripes)
Example: Eight 500GB drives in a RAID 60 gives a usable capacity of 2TB (2000GB) (RAID 6 of four 1GB stripes)
Just a Bunch of Drives refers to a collection of drives which are either not configured in a RAID - so the individual drives are accessible as separate volumes - or it can refer to storage array units which have no inbuilt RAID functionality, but are designed to connect to an upstream RAID controller. For example, you may have a SAN array which will have an inbuilt RAID controller. Capacity can be expanded on some models by adding a JBOD array. The JBOD has no RAID functionality but the drives will seen, managed and incorporated into a RAID configuration by the main controller in the upstream SAN array.
As above, RAID6 gives an additional level of protection against drive failures, due to the extra copy of parity information. However, this can come at the cost of performance as RAID6 places a greater load on the RAID controller.
Using a powerful hardware RAID controller (such as the second generation SAS 6G controllers) can provide good RAID6 performance and this is situation RAID6 is strongly recommended. Some customers have historically used RAID5 plus a hot-spare disk and this can be turned into a reliable RAID6 setup without loss of capacity or increase in cost.
Applies to: