Using AVM with RAID10

OK...just I understand.

If I had 40 disks, would I create 20 - RAID1 RAID Groups and then carve each RAID group into 2 LUNs and present these LUNs

to the Celerra?

Thanks for the quick reply

Chris

Hi,

Celerra only supports Raid 1/0 with 2 disks, and usually create 2 LUNs on each Raid Group.

Gustavo Barreto.

Yes...exactly.

Regards,

I would suggest to just use the Storage Provisioning Wizard

Thanks Rainer,

I typically do not use that wizard but I did click through it and you still

need to know what you want to accomplish to use this wizard.

Would you agree in order to use RAID1/0 in the Celerra with

40 drives you would create 20 - 2 drive RAID groups and then

create 2 LUNs per RAID group and then assign these to the

Celerra?

Thanks for your input

Chris

Yes

And thats exactly what the wizard will configure for you if you tell it you want max protection

Like with the clar_r5_performance using AVM, it will be looking for 4 LUNs to stripe.

Assuming I am right on this point...

If I use 40 drives and set to max protection and it creates 20 - 2 drive RAID groups,

would it not just use the first 8 drives?

My goal would be to have AVM leverage as many spindles as each file

system is created.

I apologize for so many followups 

Chris

SPW just creates the raidgroups and LUNs and makes them visible to the Celerra

how they are used later is up to AVM or MVM

I you want more disks involved than the AVM default for that profile I suggest creating the stripes yourself and putting them into a user defined storage pool.

This way you get both - control over layout and performance - and the ease-of-use of pools

Rainer

Gustavo: I believe that since 5.6.46 or so raid10 is supported on some platforms.

I'm interested in this as well as we have a large ns80 that is entirely raid1, and we're seeing serious disk performance issues on our busiest filesystems.

I have been doing some experimentation with user-defined AVM

and I believe what you are suggesting maybe best.  The issue

I see is limited striped members (max 8) and the Celerra metavolumes are

not striped.

I want to make sure the Celerra will not do anything unusual

with the dvolumes that would affect performance.

Want to use 20 drives in a RAID 1/0 to optimize performance,

2 - 10 drive R1/0 RGs

Create 2 LUNs on each RG (LUN 1 on SPA from 1 RG amd LUN 2 on SPB from the other RG)

Create a striped metalun of LUN 1 and LUN 2 (MetaLUN1)

Present MetaLUN1 to the Celerra and it will assign a dvolume number.

Create a user-defined storage pool with this dvolume.

I should not need to worry about template, num_stripe_members, or stripe size since

I have already laid out the disks the way I want.

The only option I should want is to "slice by default"

Now that the disk is presented and in a pool I can create a file system.

As I need to grow, I can follow this same process that I used for MetaLUN1 and just extend that

user defined storage pool (which will do a concatenation) with another dvolume using the command

nas_pool -name -xtend -volumes

Do you see any issue with the following configuration?

Are there any options or parameters in the Celerra that maybe set by

default when I create the storage pool that would cause issues based on this layout?

I appreciate the help and I am looking to optimize

performance with the drives I have available is the objective

Thanks

Chris

Hi Chris,

please do NOT use Clariion MetaLUNs for Celerra - they are not supported.

Also with RAID10 you need to stick to 2 drive raidgroups and stripe one the Celerra side

I would create 10 2-disk R1 raidgroups with one LUN each and alternating SP ownership

then create a stripe volume on the Celerra and put that into a user-defined pool

Rainer

Thanks Rainer for being so responsive and I am glad

I asked.  I do not want to do anything that is unsupported.

To get back to basics, there are 3 types of configuration

as I understand it.

system AVM

user defined AVM

user defined MVM

I have played with the user defined MVM where

I presented the LUNs as you suggest.

20 drives with 10 R1/0 RGs

2 LUNs per RG with each LUN per

RG on a different SP. (total 20 LUNs)

I presented all 20 LUNs (10 on SPA/10 on SPB)

to the Celerra.  I then manually created 2 stripes

using 10 LUNs from different RGs (5 LUNs on SPA/5 LUNs on SPB)

I then created a metavolume using the Celerra Manager where the

2 stripes are concatenated and this process created a storage pool.

So I now have a storage pool as follows

storage group (metavolume) = stripe1+stripe2

As we want to grow, how would I add a 3rd or more stripes to this

metavolume/storage pool?

After this storage pool is created, I know I can click extend

but it just asks me by how much MBs do I want to expand by...

Will it know automatically to pull in that 3rd or more stripes I created

and add extend the existing metavolume?

How do I get that 3rd or more created stripe to be part of this storage group/metavolume?

Thanks again for your help

Chris

Hi Chris,

I am not sure if you are really using a user-defined pool – last time I did that 5.6 the pool needed to be created manually.

Then you add volumes to it. Not sure how it is called in the GUI these days – I just use nas_pool and nas_volume

My suggestion to just create one LUN is because the normal default of two LUNs is mainly so that we are always sure its SP balanced and we get a number of LUNs (I/O qeues) to work with.

Since you have “small” LUNs with RAID1 just creating one LUN per RG can save some head movement.

In terms of stripes I would make that dependent on how you later plan to extend the pool. If you think you are adding disks in multiples of 20 disks then go with a stipe containing 20 disks. If you think you will rather add only 10 then use 10

No need to concat before putting storage into the pool – a pool does that automatically when needed.

For extension just create new stripes and add the stripes to the pool

Rainer

Rainer :

I'm curious why you're saying he has to stick to 2-drive raid groups? The docs imply that 2, 4, 6, or 8 disk raid groups are supported in 5.6.44+

Considering the performance problems we're suffering with raid1 mirrored pairs, I'd like to see people understand that better options are available.