[clue-talk] BAARF - Battle Against Any Raid Five (er, Four, er Free.)

[Nate Duehr]

On Oct 21, 2007, at 10:35 PM, Collins Richey wrote:

> On 10/21/07, Dan Poler <dpoler at redhat.com> wrote:
>
> [ much useful data snipped ]
>
>> Bottom line is, don't assume RAID5 is evil; depends on the need  
>> and the
>> hardware and software driving it. If you need to deploy a storage
>> solution, spend some time playing around with tools like iozone to  
>> see
>> what makes the most sense for your needs -- configure a couple  
>> ways and
>> do performance testing that mimics the block sizes your  
>> application (or
>> filesystem) will be using.
>>
>
> Finally someone with a lick of common sense.

That's pretty rude, Collins -- not like you.  Apparently you had  
nothing to add to the conversation other than an insult?

You apparently don't believe I would have recommended the same  
thing?  Block size performance and RAID performance aren't (even  
close) to the same topic.  They're certainly related when looking at  
the overall big picture of a storage solution, but this is a  
discussion of the misuses of RAID-5, not a complete "solution"  
discussion.

I was pointing out that few people understand the basic differences  
between the different RAID levels, and don't take those basic design  
differences into account when they want a redundant drive array.   
This started by sharing a website with some halfway-decent technical  
articles for the reasons why RAID-5 isn't always the correct solution.

Some people pick "RAID 5" for their md arrays because they've heard  
of it before, and "5" is a higher number, so it just *must* be better  
than a measley little "1", right!?! - Ha -- and then they go with it  
-- without knowing they could up their performance AND their  
reliability both, by using 01/10 in many cases.  Each case has to be  
looked at separately.

For small servers doing all of this via "md" with no help from the  
hardware (RAID + hardware cache in hardware), the use of RAID 5 in  
certain scenarios can be a severe performance hit, and also turns out  
to be a reliability hit, when the risk-analysis math is done.

That's all I was saying.  I'm not wrong.  Feel free to explain your  
position though, since you have one (apparently) but didn't bother to  
share.

--
Nate Duehr
nate at natetech.com