Apologies as above I clearly cannot get two raid 1/0 (8+8) per DAE as that is 16 disks !


Effectively I would need raid 1/0 (10+10) disks 0-9 bound across enclosure 1 and 2 and raid 1/0 disks 10-13 (2+2)again across enclosure 1 and 2 and create a metalun across the two raid groups ?

The final disks in position 15 would then be assigned as hot spares

Hi Storm,

First off, the largest RAID 1/0 group you can create is 8+8.
So to use all your 30 drives, you will have to have multiple RAID groups.

Secondly the only RAIDgroup size cautions refer to RAID5/RAID6, in "Number of drives per RAIDgroup (page 37 of R26 BPG)" it specifically refers to RAID 5 in high bandwidth workloads.

So, for RAID 1/0, find a group size that splits your usable (total - hot spare) drive count evenly, as it is best to metaLUN actoss RAIDgroups comprised of some equivalent number of drives. RAID 1/0 & striped metas are fairly forgiving and there will be no noticeable effect if you have some that are 4+4 and others 3+3. My rule for RAID 1/0 group size is: whatever works best with the client's logistical requirements.

Note we suggest splitting your RAIDgroups between two DAE if on different buses for a slightly higher availability, but that is not critical either.

Regards,

Dave Z

Thanks for your input Dave,

Effectively to get the required I/O level then, I will need to create a raid group 1/0 comprising of 16 disks (8+8) then a further raid group comprising of 12 disks 1/0 (6+6). Both raid groups across two DAE's and on different buses

I can then create a metalun across the raid groups to give me the necessary I/O and storage required.

Bearing in mind that raid groups for metaluns ideally will be of the same number of drives, will there be any likely performance impact using the configuration described above ?

If you do striped expansion of a LUN and thus creating a metalun, each component of the meta should be equally large (number of disks), so if you need the performance of all involved disks, you should create a triped metalun and thus you need (8+8) + (8+8). If that would be too much and (6+6) + (6+6) is not enough, you could use a concattenated metalun, but the downside on that one is that performancewise the load is not equally spread accross all drives.

So: it depends on what you really need.

If you start with the performance requirements - what type of disks are the new one?

10K FC Disk = 140 IOPS
15K FC Disk = 180 IOPS

DISK IOPS = 2500

10K Disks = 2500 / 140 = 17.8 or 18 total data disks
15K Disks = 2500 / 180 = 13.8 or 14 total data disks

Remeber to check page 45-47 in the Best Practices (v26) guide to refine this to your specific IO Load.

regards,

glen kelley

Hi
Check with your local emc SE, but you might want to add the disk in a specific order when you create your raid group so the primary disk alternate from one bus to the next.
primary is a 1 and secondary is 2
example
bus three 21212121
bus two 12121212

bus one 21212121
bus zero 12121212

I was told this provide better performance by a EMC performance guy, but it was about four years ago.

Isnt the point of spliting a raid 10 over enclosures for resilence? Surely if you're going to the expense of mirroring disk at least put source and primary mirrors on separate enclosures that way the raid group would be resilient to an enclosure failure. I far as I am aware the default through navisphere does not allow you to do this and can only be performed manually via the cli. Equally splitting primary and secondary across over different busses will probably improve performance and may provide extra resilience against bus failure as well.

AFAIK know there's little or no performance gain nowadays, but a healthy mind would suggest to do so indeed

True, however I've never encountered EMC in doing so. But you indeed need the CLI to group the disks in the right order.

Welcome to the forums ! Nice nick by the way.... Clariion with 2 r's !

the clariion with 2 r's is because I can never ever spell clariion correctly without a spellchecker.

It is one of the most annoying words to type I've come across and I have to use it almost daily.

Funnily enough this forums spellchecker doesnt recognise the word clariion