Kbase 20706: The Final Word from development on RAID 5
| Autor |
Progress Software Corporation - Progress |
| Acesso |
Público |
| Publicação |
29/07/2005 |
|
Status: Verified
GOAL:
Why RAID 5 is not recommended.
GOAL:
Final Word from Development on RAID 5
FIX:
The various RAID configurations are all different. Each configuration has specific strengths and weakness that often trade off performance, reliability, cost, etc.
- For database servers, you should optimize for maximum I/O capability. Not cost, not storage capacity. You want as much throughput as possible with as many spindles as possible.
- RAID 5 SHOULD NEVER BE USED for database systems. Period. No exceptions. No excuses. Does not matter who the vendor is or what they say.
- RAID 5 is optimized for the WRONG thing: cost and storage capacity. But disks are cheap. Very cheap. The storage vendors would make more money by not advocating RAID 5.
- RAID 5 does not provide good write performance. It is inherent in the design.
- RAID 5 is /terrible/ when you have to recover from a disk failure because to recover one failed disk, you must read the entire contents of all other disks.
If using RAID, you should ALWAYS use RAID 1+0 (mirroring and striping).
This is very black and white. There is NO room for discussion.