RAID5 gets a bad rap

[Bill Davidsen]

Gordon Messmer wrote:
> Bill Davidsen wrote:
>> Gordon Messmer wrote:
>>> Your assertion ignores the fact that filesystems themselves are, in 
>>> fact, databases.  Real-world experience with many production systems 
>>> and many workloads has convinced me to use RAID 5 as rarely as 
>>> possible. Even when I'm forced to use it, I generally choose a RAID 
>>> 5+0 configuration as I get much better performance.
>>>
>> Or you might want to read the man pages for md and mdadm. RAID10 is 
>> faster (assuming you use the "far 2" config). No, RAID10 is not 
>> another name for RAID1+0...
> 
> 
> When I read the man page for "md", I see:
> 
>    RAID10
>        RAID10  provides  a combination of RAID1 and RAID0, and is
>        sometimes known as RAID1+0.
> 
> ...so I'm not sure what man page you've been reading.
> 
The rest of that section... the md implementation is not a stripe over mirrors, 
but overlays the mirroring over all devices. The first paragraph is confusing if 
you stop there (and maybe if you don't). But there is no array of RAID0 striped 
over multiple arrays of RAIDx (x =1,5,6), it's all one array, and significantly 
faster than 1+0 in the case where the layout is far=2.

> I know that RAID 10 is faster than RAID 5+0.  I meant that in the rare 
> circumstance when I'm trying to build a very large volume of disks (say, 
> 20), I'll usually create four RAID5 arrays with five disks each, and 
> then stripe them.  The resulting array (RAID 5+0) will be more resilient 
> to failure and perform much better than a single RAID5 array containing 
> all 20 disks.
> 


-- 
Bill Davidsen <davidsen at tmr.com>
   "We have more to fear from the bungling of the incompetent than from
the machinations of the wicked."  - from Slashdot