RAID - at many levels
Posted in General, Storage Interconnects & RAID, Advisor - Steve by Steve RogersI was asked a question about which raid level(s) to use from someone who was upgrading their server at a 30 person CPA firm. One IT consultant recommended using raid “1″ for the operating system, and 4 discs in raid “10″ for data.; vs. just a 4 disc raid “10″ setup for all their OS and data.
Not knowing the application and business/application drive failure tolerance, I’d side with the consultant.
In general, it is a good practice to mirror your operating system. Segmenting raid levels for different types of data is also a common practice; especially for separating operating system data from user data.
Relative to raid 10 , it is considered one of the highest on the availability scale (refer to my earlier blog entry on which RAID is right for you, ) it is also one of the least efficient in disk utilization, as the useable capacity is only 50% of your total available drives. It’s drive failure tolerence in a 4 drive configuration would be equivalent to two raid 1, raid sets.
Raid 10 is typically used with mission critical database applications where you absolutely MUST have a mirror of your data, and want the absolute best performance the array can deliver.
Lastly…. most, if not all, RAID controllers can handle multiple raid sets on the same controller and can slice and dice the available disk drives accodingly.
Steve
July 24th, 2006 at 1:01 pm
In reading this and other postings regarding RAID utilizing SAS components, the question comes to mind: Is there any longer an advantage to separating the mirrored drives in a RAID 10 array by SCSI channel? Is not each SAS drive effectively on it’s own channel?
July 25th, 2006 at 10:19 am
Yes, SAS is a point-to-point architecture.
Having separte and distinct RAID sets for your Operating System and User Data sets is a good practice because it:
1. Allows independent disk IO to occur to each RAID set, taking advantage of available bandwidth.
2. Isolates the OS from the data from a failure perspective
3. In general, most disks do really well with sequential IO, and to a lesser degree, are slower with Random IO. In general, sequential IO from multiple applications accessing the same Volumes/RAID sets can cause the access patterns to the disks to become more like random IO, affecting your overall performance. Having separate RAID sets helps to match the disks performance more closely to the applications access needs. RAID controllers compensate for this somewhat, but in the case of JBODS, this scenario becomes more pronounced.
Hope this helps..