Difference between "RAID groups for disks" and "RAID groups for array LUNs"

[daniel_kaiser]

Hi All,

I'm reading the section "Considerations for sizing RAID groups for disks" and Considerations for Data ONTAP RAID groups for array LUNS" in the 8.0. 7-Mode storage Management Guide (smg.pdf), pg 103, and not quite sure what it means by "RAID groups for array LUNS."

What does this mean:

------------------------------------

Follow these steps when planning your Data ONTAP RAID groups for array LUNs:
1. Plan the size of the aggregate that best meets your data needs.
2. Plan the number and size of RAID groups that you need for the size of the aggregate.
Follow these guidelines:
• RAID groups in the same aggregate should be the same size with the same number of LUNs
in each RAID group. For example, you should create four RAID groups of 8 LUNs each, not
three RAID groups of 8 LUNs and one RAID group of 6 LUNs.
• Use the default RAID group size for array LUNs, if possible. The default RAID group size is
adequate for most organizations.
Note: The default RAID group size is different for array LUNs and disks.
3. Plan the size of the LUNs that you need in your RAID groups.
• To avoid a performance penalty, all array LUNs in a particular RAID group should be the
same size.
• The LUNs should be the same size in all RAID groups in the aggregate.
4. Ask the storage array administrator to create the number of LUNs of the size you need for the
aggregate.
The LUNs should be optimized for performance, according to the instructions in the storage array
vendor documentation.

----------------------------------------

We're purchasing a new 3240 cluster array with 2x DS4243 w/600GB SAS drives and 1x DS4243 w/1TB SATA and looking to use 8.0.1 with 32-bit aggregates (snapmirroring to another filer with 32 bit aggregates.  For a DS4243 shelf with 600GB drives, what is recommended.  In the Storage Management Guide it states that you should add the max disks in a raid group, which is 28 disks for performance.  Is this the right path to take?

[peter_lehmann · Accepted solution]

Hi Daniel

Answering the question in the title first:

"RAID groups for disks" is related to creating an aggregate with "NetApp Disks" in a NetApp shelf.

"RAID groups for array LUNs" is related to "Disks" in a Disksystem form any other vendor (HDS, EMC, HP etc...)

To your question related to the raidgroups:

What it means is, that when creating the aggregate, one should take care to create raidgroups of equal size .This way you make it easier for WAFL to properlylayout the stripes and guarantee performance.

As an example, when you have on shelf DS4243 with 600GB SAS Disks (use 24 disks in order to get the optimum layout, assuming you have sufficient spare somewhere. If not, then use 21 disks (for 3 raidgroups) or 22 disk for 2 raidgroups), examples follow:

using 24 disks and 3 raidgroups:

filer> aggr create aggr1 -B 32 -r 8 24

this creates the aggr1 with 3 raidgroups, 8 disks each.

using 24 disks and 2 raidgroups:

filer> aggr create aggr1 -B 32 -r 12 24

this creates the aggr1 with 2 raidgroups, 12 disks each.

using 21 disks and 3 raidgroups:

filer> aggr create aggr1 -B 32 -r 7 21

this creates the aggr1 with 3 raidgroups, 7 disks each.

using 22 disks and 2 raidgroups:

filer> aggr create aggr1 -B 32 -r 11 22

this creates the aggr1 with 2 raidgroups, 11 disks each.

Just try to make sure to have equly sized raidgroups in your aggregates.

Of course you can create the aggr1 with 24 (or 28) disks, with just one raidgroup. Something I see all the time and works just fine, if yor are aware of the higher risk of loss of data (which is more or less mute with RAID-DP)... However, NetApp Best Practices are not stating this, because they do not want to have the lawyers turn up and try to make money with this. Fact is: the aggr with 2 or 3 smaller raidgroups is bettter protected then with one big raidgroup.

Hope this helps,

Peter

[daniel_kaiser]

Thanks Peter for the detailed explaination.  Since i was mainly reading the Disk Group section of the SMG guide, I didn't realize that they talked about vfilers and LUNS in the previous section.

One more question.  Since we have a DS4243, would i specify 2 drives as hot spares, which leaves me with 22 drives, and then create two raid groups (11 drives each)?  If i went this route, I would only have about 16 "data drives" as 4 drives would be parity drives and 2 drives would be hot spares.  16 x 600 = 9.6TB raw, but realisticly, after you rightsize the drive and consider wafl over head, the drive sizes are more like ~410GB.  If this is the case then 16 x 410 is roughly 6.5TB.  Since we only have 2 shelves of DS4243, i'm really debating between setting up 2 raid groups for each shelf or 1 raid group of 22 drives per shelf.  Any additional input is more than welcome.

Okay last question.  Are hot spares assigned to a controller or can they be shared between controllers.  We have an 3240HA and looking to assign disks in one shelf to one controller while the other shelf to the other controller.

[peter_lehmann]

No problem. It is relatively easy to get confused these days, since NetApp shelfs are supported on V-Series too.

One is always forced to decide between 1) maximum space  2) maximum security ... And in 99% of the cases option 1 wins

Maximum space equals to: use all the disks "except one" and set the raidgroup size to 28 (or 23).

For Hot Spares... They are global for each controller, which means you will have to have at least one Hot Spare per HA controller.

This will leave you probably with this setup:

Controller-A                      Controller-B

Shelf1                                 Shelf1

aggr with 23 disks              aggr with 23 disks

1 Hot Spare                        1 Hot Spare

Of course there a million more versions to this, but this is waht it most often comes down to.

Hope this helps,

Peter