RAID Systems (page 2)
There were 6 RAID levels defined.
Note: The numbers used to describe the RAID levels do not imply improves performance, complexity, or reliability.
- RAID-0 is Striping RAID. It does not, by itself, contribute to EDAP and provides NO redundancy, since if one drive fails, all the data in the array will be lost.
- Two types of RAID provide EDAP for the drives: Mirroring and Parity. Mirroring appeared earlier (in the UC Berkeley Papers) and was originally designated as RAID Level 1. Its main problem is that it requires 100% redundancy or twice as much capacity as was originally needed. On the other hand, its read performance is improved and a higher percentage of drives in a RAID-1 system may fail simultaneously as compared to a Parity RAID system.
- RAID Level 2 requires the use of non-standard disk drives and is therefore not commercially viable.
- Parity RAID was identified in the UC Berkeley Papers as RAID Levels 3, 4, 5 and 6. It RAID significantly reduces redundancy overhead to a range of 10% to 33% (compare to 100% for Mirroring RAID-1). RAID Levels 3-5 provide EDAP in case of one disk failure, and RAID-6 tolerates failure of two disks at the same time or if the second disk fails later, during the reconstruction period.
Levels 1, 3, and 5 are the most commonly used RAID levels and are discussed in the next table. Lets consider a number of drives, say, N drives, connected to the server with MTBF equal to M, and with mean time to repair R.
RAID-0
- Provides NO redundancy, since the data are written across multiple drives (so-called stripping). If one drive fails, all the data in the array will be lost.
- Provides higher data rates, since all drives are accessed in parallel.
RAID-1
- Data mirroring. High reliability. The same data is written or read on two (or more) drives.
- Faster reading, since the first drive to respond to a request will provide data, thus reducing latency.
- The coast at least doubles for a given storage capacity.
- * MTBF ~ 2M + M2/R (see legend below)
RAID-3
- One extra drive is added to store the parity data (error correction data). If one drive fails, the data can be recovered and the other drives will keep working till the failed one is replaced (of cause, performance will suffer).
- High reliability (cheaper than mirroring in RAID-1).
- Very high data rates. Data writing and reading occurs in parallel.
- For a given capacity, fewer drives are needed than for RAID-1.
- Controller may be more complex and expensive.
RAID-5
- Data and parity information stripping across all drives.
- High reliability, high performance
* Lets consider a number of drives, say, N drives, connected to the server with MTBF equal to M, and with mean time to repair R.
Data Availability
RAID-0 is inappropriate when data availability (reliability) is an issue, since it provides no data redundancy and failure tolerance. Data mirroring (RAID-1) or parity check (RAID-3 to RAID-5) are needed in this case, leaving. RAID-5 is recommendable because of its combination of high data availability and good performance. In case of drive failure, even if the data is still physically available, it may be inaccessible within the needed period due to a drastic drop in system's performance. Performance drop of, say, 50% is not totally unusual during reconstruction of the system.
Back to Top