Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

544

May 23rd, 2008 06:00

RAID Configuration

Hello,
Most, if not all, of the whitepapers I've read regarding RAID levels in a clariiion mention 4+1 or 8+1. Is there a reason why 5 or 9 disks are recommended for a RAID5 configuration? The reason I ask is because we currenlty have a 4+4 configuration, and we're thinking of implementing TWO 3+1 and use MetaLUNs from those spindles. I have not seen any whitepapers that mention using 4 disks for RAID5s. Any ideas?

thanks,

j

2 Intern

 • 

5.7K Posts

May 26th, 2008 02:00

The assumption is correct. As long as you don't need "more performance and less capacity" I would definately go for the "somewhat less performance and more capacity". It makes no sence having less capacity but lots of performance which you're not going to use.
Another issue to think about is that 2 (3+1) LUNs bound together as 1 metalun only can survive the loss of 2 disk failures (1 in each RAID Group), while in the 4+4 set up, each mirrored pair can survive a disk failure.

If you use up all the space on the RG's and the performance is still not the bottleneck and the future does not show any performance issues, you're good to go.

So you see: it all depends on your needs / wishes.

133 Posts

May 23rd, 2008 07:00

Hello jjimenez and welcome to the forums!

R5 with 4+1 and 8+1 are often mentioned but there is no hard and fast rule about having to only stick to those RG sizes. Often though they are seen as the sweet spots when thinking about for performance, RAID overhead and rebuild times etc etc.

If you are going to implement MetaLUNs with R5 the recommendation as per FLARE Release Best Practise guide is to stick with a minimum of 3+1.

But in reality what gets implemented as always depends on available spindles, IO and capacity requirements.

A question for you though, what are you hoping to achieve by changing the RAID from a R10 4+4 to R5 2x3+1 as the IO is still distributed over exactly same amount of spindles? Secondly what issues (performance?) are you experiencing and what application(s) are you using?

I hope this helps
Carl

2 Intern

 • 

5.7K Posts

May 23rd, 2008 07:00

For RAID5 anything from 3 to 16 disks is supported.
For RAID3 only 4+1 and 8+1 are supported.

So in R5 you can also have 6+1 or 11+1.

the 4+4 you mention is RAID10.

Metaluns are good as long as the components of the meta are on different raid groups. You can choose between striped meta's, which is good for performance, since all data gets evenly spread accross over all disks involved.
Concattenated meta's are easy to use, since the additional space for a lun is instantly available while with striped meta's you need to wait until the restriping is done.

103 Posts

May 23rd, 2008 11:00

Thanks for your response RRR.

Carl,

Well, we're currently using our environment for VI3. Originally, we (Dell, VMware, and us) created a R10 for our "Data" on virtual machines. After doing some research, I've noticed that going with 2xR5 has "somewhat" the same performance as a R10. And not only that, we'll have more capacity with the 2XR5 setup. I don't think we're having performance issue, at least that I know of. My thought was, why not implement something that has similar performance but has higher capacity.

Is this a correct assumption?

133 Posts

May 27th, 2008 03:00

Jjimenez,

RRR has already answered the question yesterday whilst we were enjoying a National day off work here in the UK. One final note though, if you need to further expand you MetaLUNs you should do that by creating another R5 3+1 and a component LUN within it of same size as your current LUNs. It should also be striped to keep the performance as predictable as possible.

I hope this helps
Carl
No Events found!

Top