Comments on: SQL SERVER – Raid Configuration – RAID 10

By: Travis Dean

Travis Dean — Tue, 28 Apr 2009 18:39:14 +0000

This is directly from that same article if you look below it:

RAID Level 10 Explanation Contains an Error

The RAID Level 10 explanation given here in the MS documentation is actually for Level 0+1. Level 10 is striping the mirrored segments of disks.

For Example: If you have 4 drives, drive 1 and 2 are mirrored, drive 3 and 4 are mirrored having 2 logical drives, then stripe them. Here you have high fault tolerance and excellent performance because of striping.

By: pinaldave

pinaldave — Tue, 21 Apr 2009 03:08:11 +0000

Andrew Morgan,

I am absolutely very much interested in what your test results are.

Kind Regards,
Pinal

By: Andrew Morgan

Andrew Morgan — Tue, 21 Apr 2009 01:35:18 +0000

Pinal,

This is going to be an interesting case study. We currently use disk mirroring (RAID 1) and are completely unimpressed with our write performance. We are going to be switching to RAID 10 shortly. Look forward to letting you know about our performance increase!

By: Xwris Onoma

Xwris Onoma — Sun, 15 Mar 2009 06:29:12 +0000

Yes! RAID-10 is first RAID-1 then RAID-0. RAID-01 (not to be confused with RAID-1) is first RAID-0 then RAID-1. Please do a google search on these terms and spend up to 1/2 hour reading up. No IT professional should ever say that “RAID-10 is 0+1″. Yes in plain arithmetic 0+1=1+0. But this is not the same. Be responsible people! Others are reading you too!

By: SQLAuthority News - Best Articles on SQLAuthority.com Journey to SQL Authority with Pinal Dave

Tue, 24 Feb 2009 12:07:43 +0000

[...] SQL SERVER – Raid Configuration – RAID 10 [...]

By: Keld Simonsen

Keld Simonsen — Wed, 21 Jan 2009 23:45:41 +0000

You are right, what is commonly called rai10 is actually raid1+0, and raid01 is actually raid0+1.

But there is a raid10 type in Linux, which is different, it is not a nested raid type, but all are done within the same kernel driver. It has the advantage of doing striping on all constituing drives at sequential reads, with the “far” layout, giving almost the performance of raid0 . It is well suited for SQL operations, as there is also some optimizations in using only the outer faster sectors for reading, and also limiting read seeking to the outer (faster) sectors on the disk..Raid10 can be constructed with only 2 drives.

More on linux raid incl raid10 can be found on http://linux-raid.osdl.org

raid5 cab also be very fast, for sequential reading.with an observed performance ov a little over N-1 times the drive nominal speed, N being the number of drives active. This actually also goes for sequential writing, having a write speed of N-1.

For SQL uses RAID5 is not the best choice, the writing speed is something like N/4 because you need 2 reads and 2 wrtes to change a smaller data block.

By: subash

subash — Wed, 14 May 2008 15:45:31 +0000

Hi, i have been going thru the article on RAID. I found that it is defined as Redundant Array of Inexpensive Disk. While it may sound correct, isn’t the real definition is Redundant Array of Independent Disk?

By: Krishnaraj

Krishnaraj — Tue, 26 Feb 2008 19:14:24 +0000

I agree with Rich.

RAID 01 is a mirrored configuration of two striped sets.
RAID 10 is a stripe across a number of mirrored sets.

Both will combine the Striping and Mirroring But the Difference is which level applied first.

By: Rich

Rich — Fri, 25 Jan 2008 17:39:16 +0000

Alan and Dan are correct. Pinaldave and Paul are incorrect. RAID 0+1 is NOT the same as RAID 10. The RAIDs differ because the RAID modes are layered (nested) in the opposite order. RAID 0+1 is vulnerable to single drive failure, as Alan observed correctly. This point is confirmed and reinforced on multiple RAID reference sites.

By: cheng

cheng — Wed, 23 Jan 2008 16:22:29 +0000

I tried to shrink production db it has three files of data. each of them left about 220 MB free space.
database data_1 size 2192 MB, free space 0 MB.
data_2 size 2219 mb, free space 224.6 mb
how can I shrink them?

DBCC shrink database (proDB, ?)
Go

By: Paul

Paul — Fri, 18 Jan 2008 17:57:13 +0000

Sorry Dan but you are incorrect. The industry says that RAID 0+1 is also known as RAID 10. I am 100% certain

By: Dan Rous

Dan Rous — Fri, 10 Aug 2007 09:15:36 +0000

To quote “RAID 10 is RAID 0+1″… sorry but that’s rubbish!
RAID 10 is RAID 1+0!
RAID 01 is RAID 0+1!
So to have said “to avoid confusion RAID 0+1 = RAID 01 is called RAID 1+0 = RAID 10″ is incorrect!?
Surely you’re not claiming that RAID 0+1 is the same as RAID 10?

By: pinaldave

pinaldave — Tue, 01 May 2007 02:44:21 +0000

Alan,

It is correct. RAID 10 is RAID 0+1. As we already have RAID 1 and to avoid confusion RAID 0+1 = RAID 01 is called RAID 1+0 = RAID 10.

Your observation about RAID is correct. Thank you,

Pinal Dave
(http://www.SQLAuthority.com)

By: Alan Saunders

Alan Saunders — Mon, 30 Apr 2007 16:08:11 +0000

Surely the RAID array described in the ‘Book online’ link above is RAID 0+1 not RAID 10 (1+0)
RAID 0 (striped) Mirrored to another RAID 0 i.e stripe two sets of disks then mirror the stripes

RAID 10 as I understand it is a RAID 0 Array of RAID 1 Mirrored pairs. i.e. Mirror pairs of disks, then stripe the mirrored pairs.

In RAID 0+1, any one drive failing from each RAID-0 array (2 drives) will fail the entire array, whereas any two drives failing in a RAID 1 + 0 array will only fail the entire array if the drives concerned are both halves of the same mirrored pair.

Please tell me if I’m wrong